Waveshare ESP32-C5 Pico board brings the ESP32-C5 dual-band WiFi 6, Bluetooth LE, and 802.15.4 (Zigbee/Thread) SoC to the Raspberry Pi Pico form factor.

The board features an ESP32-C5-MINI-1 Series module with PCB or external antenna, two 20-pin GPIO headers, a USB-C port for power and programming, Reset and Boot buttons, and a few LEDs. It also supports Li-Ion battery charging and discharging.

Waveshare ESP32-C5 Pico specifications:

• Wireless module – ESP32-C5-MINI-1 or ESP32-C5-MINI-1U
  • Wireless MCU –  Espressif Systems ESP32-C5HF4
    • CPU
      • Single-core 32-bit RISC-V processor @ up to 240 MHz
      • Low-power RISC-V core @ 48 MHz acting as the main processor for power-sensitive applications
    • Memory – 384 KB SRAM on-chip
    • Storage – 320 KB ROM, 4MB SPI flash
    • Wireless
      • Dual-band (2.4/5 GHz) 802.11ax WiFi 6, 802.11b/g/n WiFi 4 backward compatibility
      • WiFi modes: Station mode, SoftAP mode, SoftAP + Station mode, and promiscuous mode
      • Bluetooth 5.0 Low Energy (LE) with Mesh support, up to 2Mbps data rate
      • 802.15.4 radio for Zigbee 3.0, Thread 1.3, and Matter up to 250 K
  • ESP32-C5-MINI-1 – PCB Antenna
  • ESP32-C5-MINI-1U – IPEX antenna connector (antenna provided with the kit)
• USB – USB Type-C port for power/charging and programming
• Expansion
  • 2x 20-pin Raspberry Pi Pico-compatible headers with up to 26x GPIO, 20x PWM, I2C, SPI, UART, SDIO, JTAG, Serial, 4x ADC, VBUS, VSYS, VBAT, 3.3V, GND
  • Note: 6x GPIO pins through TCA9554PWR GPIO expander
• Misc
  • Boot and Reset buttons
  • Power, Charging, and reverse polarity warning LEDs
  • RGB LED
• Power Supply
  • 5V via USB-C port or VBUS pin
  • 3.7V Lithium battery support via 2-pin battery connector
  • MP1605GTF-Z synchronous buck (step-down) DC-DC converter taking 2.3–5.5V input
  • Eta Solutions (Yutai Semiconductor) ETA6098 switching Li-Ion battery charger
• Dimensions
  • ESP32-C5-MINI-1 variant – 56.85 x 21 mm
  • ESP32-C5-MINI-1U variant – 51 x 21 mm

Four variants of the board are offered:

• ESP32-C5-Pico – PCB antenna, unsoldered headers
• ESP32-C5-Pico-M – PCB antenna, pre-soldered headers
• ESP32-C5-Pico-U – External antenna, unsoldered headers
• ESP32-C5-Pico-UM – External antenna, pre-soldered headers

The company provides support for the Arduino IDE and the ESP-IDF framework. You’ll find additional technical details PDF schematics, datasheets, and instructions to get started on the documentation website.

The company used to share code through tarballs, but Arduino/ESP-IDF code samples and other resources for the ESP32-C5 Pico board can be found on GitHub instead. Code samples are available for board information, the built-in RGB, the I/O expander, dual-band WiFi, UART, and connecting a small external 1.83-inch touchscreen LCD and an SPI CAN transceiver (MCP2515-based).

Waveshare sells the ESP32-C5 Pico for $10.43 to $11.33 on AliExpress, $15.99 to $17.99 on Amazon, and $8.99 to $9.99 on the company’s online store. As usual, the prices don’t include shipping and taxes since these vary by country.

The post ESP32-C5 Pico board follows Raspberry Pi Pico form factor, ships with on-board or external antenna appeared first on CNX Software - Embedded Systems News.

The FREE-WILi 2 (Founder Edition) is an open-hardware, portable hacking multitool and lab designed for hardware hackers, pentesters, and embedded system developers. Designed as an “open electronics multitool,” it combines a range of chips and modules into a handheld device, including two Raspberry Pi RP2350 MCUs, an ESP32-C5 for wireless connectivity, a Lattice ICE40UP5K FPGA, and a Raspberry Pi CM0 for running a full Linux OS.

The device also features a 3.5-inch capacitive touchscreen, a 14-button physical interface, and modular “Orca” headers for hardware expansion. Its wireless features include 2.4/5GHz Wi-Fi, Bluetooth, LoRa, sub-GHz RF, NFC, 125kHz RFID, and IR. These features make it very suitable for wireless testing, embedded hardware debugging, real-time control, electronics development, and retro gaming, with support for Adafruit Fruit Jam, PICO-8, and Doom.

FREE-WILi 2 specifications:

• Compute
  • Main MCUs – 2x Raspberry Pi RP2350B (One for main controls, one dedicated to display)
  • SoM – Raspberry Pi CM0 (to run a full 64-bit Linux OS and run heavy Python scripts)
    • SoC – Broadcom BCM2710A1
      • CPU – Quad-core Cortex-A53 processor @ 1.0 GHz
      • GPU – VideoCore IV GPU supporting OpenGL ES 1.1, 2.0 graphics
      • VPU – H.264/MPEG-4 1080p30 video decoding, H.264 1080p30 video encoding
    • System Memory – 512MB LPDDR2 RAM
    • Storage – 8GB or 16GB eMMC flash
    • Wireless – Optional WiFi 4 and Bluetooth 5.0 module
  • FPGA – Lattice UltraPlus iCE40UP5K FPGA with 5.3K LUTs, 1Mbit SPRAM, 120Kbit DPRAM, 8x multipliers with all pins brought out; 8 MB of dedicated SRAM
• Storage – 2x MicroSD card slots (one slot is for device firmware; the other is for CM0)
• Display
  • 3.5-inch 480×320 capacitive touch screen
  • DVI output (Driven by RP2350 HSTX)
• Audio
  • 3.5 mm audio jack (headphone/mic)
  • 4-channel phased-array microphone
  • 0.5W onboard speaker
• Wireless
  • Dual-band (2.4 GHz + 5 GHz) Wi-Fi 6, Bluetooth LE, and 802.15.4 (Zigbee/Thread) via Espressif ESP32-C5
  • Sub-GHz connectivity via CC1101 module
  • LoRa via STM32WLE5JC Arm Cortex-M4 MCU @ 48 MHz with 256 KB flash memory, 64 KB SRAM, SX126x LoRa radio
  • NFC / RFID – 125 kHz RFID (LF) + ST25R3916B NFC (HF) for tag reading, cloning, and emulation
  • Infrared – IR Tx/Rx with dedicated case window
  • SMA connector for LoRa and Sub-GHz
• USB
  • 3x USB host ports (2x 12 Mbps connected to RP2350 display MCU + 1x 480 Mbps port connected to CM0)
  • 1x USB Type-C port (CLI, OneWili, GUI, high-speed SD reader)
• Sensors
  • IMU – Bosch BMI323 (9-DOF with magnetometer) + BMM350 Magnetometer for motion input
  • Environment – Sensirion SHT40 (Temperature & Humidity)
  • Light – OPT4001 ambient light sensor
• Debug
  • Raspberry Pi Debug Probe onboard (for RP2350 + LoRa debugging)
  • ESP32 JTAG via USB
• Expansion
  • 20-pin (FW1) main connector + 10-pin analog connector + mounting posts for “Orca” add-on modules
  • 13x GPIOs (SPI, I2C, UART) with software-selectable I/O voltage
  • 4x analog inputs (0–5V) with op-amp buffering, Programmable Gain Amplifier (PGA), and window comparator
  • 2x analog outputs (0 to ~4.84V @ 25 kHz)
  • CAN FD at full 8 Mbit/s with CAN SIC transceiver
• Misc
  • 5-way D-pad + A/B/X/Y (home/ok/cancel/page) buttons
  • 14 gamer buttons total (emulator-friendly for PICO-8, Doom, Fruit Jam)
  • 5x under-screen context backlit keys
  • Innovative 2-press-per-letter hardware keyboard
• Power
  • 3,000 mAh battery with USB-aware charging
  • Dedicated ultra-low-power (ULP) micro managing 17 independent power zones
  • 1–5.5V / 1.5A programmable power supply with MOSFET crowbar for voltage glitching
  • 60 µA sleep current
• Dimensions – 152.4 × 78.9 × 22.3 mm
• Weight – 290 grams

On the software side, FREE-WILi 2 runs stock firmware that exposes a USB CLI, on-device GUI, and the OneWili API for controlling the hardware from Python, Rust, or C/C++. It also supports rThon for Python-like real-time scripting, WiliBlocks for visual programming, and ZoomIO for sub-microsecond real-time control. You can also load UF2 applications from the microSD card, run WebAssembly programs through WiliWasm, and use the FREE-WILi GUI on a PC without installation.

It also ships with open hardware documentation, an open board support package (WiliBSP), and “Agent.md” files. This allows LLMs like Claude Code or local models running via LM Studio to autonomously write, compile, and debug custom firmware directly on the device using the onboard hardware debugging probes. More information is available on their GitHub account.

The company mentions that they have ported Meshtastic to the onboard STM32WLE5JC LoRa radio, so FREE-WILi 2 can join a LoRa mesh out of the box. That radio shares one external antenna with the TI CC1101 sub-GHz transceiver via dynamic switching. This is a vendor port — FREE-WILi 2 is not listed as an official Meshtastic device, and regional band settings are not documented yet.

We’ve previously written about other portable wireless hacking tools such as the Flipper Zero, MCU-based alternatives like HackBat and the M1 multitool, Linux-based devices like Interrupt and the much more powerful Flipper One; compared to all of those, the FREE-WILi 2 sits in that same pocket-lab space, but packs different types of SoCs and a wide set of radios into one handheld.

The FREE-WILi 2 Founders Edition is currently available for pre-order at $400 directly on the company’s website. Shipping is expected to begin in Q4 2026. More information is available on the project’s website.

The post FREE-WILi 2 portable hacking multitool features two RP2350 MCUs, ESP32-C5, ICE40 FPGA, and Raspberry Pi CM0 appeared first on CNX Software - Embedded Systems News.

Arm has introduced the CSS for Mobile 2 platform as an AI-native compute foundation comprised of the Arm C2-Ultra and C2-Pro CPUs with SME2 (Scalable Matrix Extension 2) units and the Mali G2-Ultra NX GPU with dedicated neural accelerators that enable neural graphics to run natively on the GPU.

Arm C2-Ultra and C2-Pro CPUs

Arm C2-Ultra specifications:

• Architecture – Armv9.3-A
• Extensions
  • Armv9.4-A (Some key features)
  • SME2 Extensions
  • SVE2 Extensions
  • Cryptography Extensions
  • RAS Extensions
• ISA support – AArch64
• Microarchitecture
  • Pipeline – Out-of-order
  • Superscalar
  • Neon and SVE2
  • Optional Cryptography Unit
  • Up to 14x CPUs in cluster
  • 40-bit Physical Addressing
• Memory System and External Interfaces
  • 64KB (I-Cache), 128KB (D-Cache)
  • 2MB or 3MB L2 Cache
  • Optional, 256KB to 32MB L3 Cache
  • ECC Support
  • Bus interfaces – AMBA AXI5 or CHI.E
  • Optional ACP
  • Optional Peripheral Port
• Security – TrustZone, Secure-EL2
• Interrupts – GIC interface, GICv4.1
• PMU – PMUv3
• CoreSight – CoreSightv3
• Embedded Trace Extension – ETEv1.0
• Trace Buffer Extension
• Optional Area-Optimised Configuration

Arm says a C2 CPU cluster delivers the following performance improvements over an equivalent C1 CPU cluster:

• Up to 1.7x performance improvement on AI models
• Up to 15 percent higher single-thread performance
• 15 percent faster web browsing
• 12 percent faster application launch
• 12 percent higher multi-thread performance at the cluster level

One practical example is lower LLM/agent latency of a dual-core C2-Ultra cluster compared to a dual-core C1-Ultra cluster, especially when SME2 units are enabled.

I could not find C2-Pro documentation at this stage (it points to C1-Pro docs), but you’ll find more technical details about the new C2-Ultra CPU core on the documentation website.

Arm Mali G2-Ultra NX GPU

Mali G2-Ultra NX key features and specifications:

• API support – OpenGL ES 3.2, Vulkan 1.4, OpenCL 3.0 full profile
• Bus interface – AMBA 4 ACE, ACE-LITE, AXI
• Configurable dual L2 cache supports up to 8 slices of 2MB each
• Scalability – 10 to 24 cores
• Adaptive Scalable Texture Compression (ASTC) with LDR and HDR support
• Arm Frame Buffer Compression (AFBC), version 1.3.2, 4×4 blocks
• Arm Fixed Rate Compression (AFRC)
• Arm Neural Technology
  • Neural Super Sampling (NSS): AI upscaling from 540p to 1080p
  • Neural Frame Rate Upscaling (NFRU): Frame rate doubling from 30 FPS to 60 FPS
  • Neural Super Sampling and Denoising (NSSD): A combination of NSS and ray reconstruction
• Ray Tracing –  RTUv3 – improved BW efficiency, higher performance, lower area, and HW-based OMM
• Variable Rate Shading – Up to 4×4 shading rate supported

Compared to the earlier Arm Mali G1-Ultra GPU, architectural improvements in the Mali G2-Ultra deliver up to 24 percent higher benchmark
performance, 14 percent higher non-AI gaming performance, and a reduction of 70% for ray tracing workloads. With neural acceleration through NFRU, the system supports longer mobile gaming sessions at up to 120 FPS, and DRAM traffic is also greatly reduced by up to 70%.

You’ll find additional information for the new GPU on the Arm website. You can also visit the launch page for more details and links about the  Arm CSS for Mobile 2 platform as a whole.

The post Arm CSS for Mobile 2 “AI-Native” platform: Arm C2-Ultra and C2-Pro CPU cores, Mali G2-Ultra NX GPU appeared first on CNX Software - Embedded Systems News.

YuzukiHD has made another open-source hardware Linux SBC. YuzukiNeko is a Raspberry Pi Pico-sized single board computer powered by a new Allwinner F101 RISC-V SoC.

The Allwinner F101 processor features a single 1.008 GHz C907 RISC-V core, 8MB or 16 MB PSRAM, and support for various display interfaces such as MIPI DSI, 24-bit RGB, and LVDS. The YuzukiNeko board itself features the F101 SoC, an SPI NOR flash, a microSD card for storage, a USB-C port, and two 20-pin GPIO headers.

YuzukiNeko specifications:

• SoC – Allwinner F101
  • CPU – T-head C907 RISC-V core up to 1.008 GHz
  • Memory – 8MB or 16MB PSRAM @ 252 MHz
  • Package – QFN-68
• Storage
  • 128 Mbit NOR flash (PY25Q128HA-SUH-IT)
  • MicroSD card slot
• USB – USB Type-C port for data, power, and programming
• Expansion
  • 2x 20-pin expansion headers with GPIOs, ADC, 5V, 3.3V, and GND
  • 5x pads with HP out (analog speaker output) and four GPIOs
• Misc – FEL button, power LED
• Power Supply – 5V via USB-C port
• Dimensions – About 51 x 21 mm

You’ll find the PDF schematics, EasyEDA PCB layout, Gerber files, 3D STEP file, BoM, and other hardware resources on GitHub. The Linux source code based on mainline (Linux 7.2) and Buildroot are in separate repositories. In addition to Linux, the board and the Allwinner F101 processor also support Zephyr RTOS.

A product brief for the Allwinner F101 “Smart Control and Display Processor” can be found on the Whycan website, so we have more details about the F101, including a block diagram and specs.

Allwinner F101 specifications:

• CPU – T-Head XuanTie C907 RISC-V CPU @ 1.008 GHz with  32 KB I-cache + 16 KB D-cache
• Display Engine
  • Allwinner Awonder1.1 Lite post-processing
  • De-interlace (DI) up to 720p @ 60fps
  • G2D hardware accelerator supporting rotate and mixer functions, among others
• Video engine
  • Image decoding
    • JPEG up to 16384×16384
    • PNG up to 2048×2048
  • Video encoding
    • JPEG/MJPEG up to 720p @ 30fps
    • Maximum resolution: 8192×8192
• Memory
  • F101S2-XXX – 8MB 16-bit PSRAM (SiP)
  • F101S3-XXX – 16MB 16-bit PSRAM (SiP)
• Storage – SPI flash interface
• Video Output
  • RGB888 up to 1280×720 @ 60fps
  • Single-link LVDS  up to 1366×768 @ 60fps
  • 4-lane MIPI DSI  up to 1280×800 @ 60fps
• Audio
  • DAC
  • Analog audio – HPOUT
  • Digital audio – I2S/PCM, OWA OUT
• USB – USB 2.0 DRD
• Other peripherals
  • SDIO2.0/3.0
  • 2x SPI, 6x UART, 3x TWI  (I2C)
  • 4-ch PWM, 2x PWM_BL, IR_RX
  • 1-ch GPADC, 4-ch TPADC
• Security – Integrated 512 bits eFuse storage space
• Package – QFN68, 7 x 7mm

The YuzukiNeko is not for sale yet, but may be soon. Other open-source hardware from YuzukiHD, like the Avaota A1 and Yuzuki Chameleon, ended up being sold online. One member on a Whycan forum thread mentions that the Allwinner F101 sells for 18 yuan, or about $ 2.70 US. This could mean the YuzukiNeko board or similar designs could sell for under $10.

Thanks to Freire for the tip.

The post YuzukiNeko – A Linux-capable Allwinner F101 RISC-V SBC with Raspberry Pi Pico form factor appeared first on CNX Software - Embedded Systems News.

When the Part Exists, but the Data Doesn’t

Reverse engineering often starts with a physical part but no reliable digital data. In restoration and customization, manual measurement can provide basic dimensions, but it becomes limiting when complex curves, mounting points, and tight clearances need to be captured accurately.

MetroY Ultra captures the actual three-dimensional geometry of a part, providing a reliable digital reference for reverse engineering, custom design, and manufacturing.

Designing a Custom Motorcycle Seat Tray from Accurate Scan Data

In a custom 1984 Honda CB400 build, the builder needed to create a seat tray and seat pan for a custom seat while integrating the rear fender into the modified rear subframe. The challenge was not simply designing a new part, but making that part accurately fit a motorcycle whose geometry had already been modified. Traditionally, this kind of work is often done with welded sheet metal or fiberglass layup. Both methods can produce strong parts, but they also require cutting, trimming, sanding, fitting, and repeated manual adjustment. When the geometry of the existing structure is not fully known, these adjustments are often part of the design process itself.

Instead of starting with metalwork or fiberglass, the builder moved the project into a digital workflow. The motorcycle had already been 3D scanned, and MetroY Ultra was then used to capture the rear fender from multiple angles, including both the top surface and underside. Markers were placed on and around the fender to support stable tracking, and enough overlap was captured so the scan data could be merged into a complete model. Capturing both sides of the fender was important because the part’s overall shape, mounting features, and relationship to the surrounding frame all needed to be represented in the digital model.

The scan was captured and initially processed in Revo Metro, Revopoint’s software for MetroY Ultra. From there, the data could be reviewed and prepared before moving into further cleanup. One practical challenge was that some marker-related holes were similar in size to the real mounting holes on the fender, so the model had to be cleaned carefully: unwanted holes and scan artifacts were removed, while actual bolt holes were preserved. This distinction was important for the downstream design: preserving the actual mounting geometry while removing scan artifacts ensured that the digital model remained a useful representation of the physical part.

Once cleaned, the rear fender scan was combined with the full motorcycle scan in CAD software. This gave the builder an accurate digital reference of how the fender, frame hoop, and seat area related to one another. Instead of working from a series of measurements and trying to reconstruct these relationships manually, the builder could now design against the actual geometry of the modified motorcycle. Using the scanned frame geometry as a guide, he designed a 3D printable seat tray with cutouts matching the actual frame, along with cable routing holes and mounting points for brass inserts.

For this type of custom-fit design, MetroY Ultra provides accurate scan data that gives the early design stage a more reliable foundation. The value is not simply having a 3D model of the fender, but having its actual geometry available as a design reference. Instead of starting with rough measurements or repeated trial fitting, the builder could test and adjust the design digitally before cutting metal, laying fiberglass, or moving toward a full-size 3D print. For a one-off custom build, this helps move fitment from a physical trial-and-error process into a digital design process.

Accurate Digitization of a Classic Car Part

A radiator grille may look like a simple exterior component, but in classic vehicle restoration it can be a fitment-critical part. In this project, MetroY Ultra was used to scan a radiator grille for a VW Golf MK1 Convertible, creating a high-accuracy digital reference for a part that may be difficult to source, repair, or reproduce.

For older vehicles, original CAD data is often unavailable, and replacement parts can vary in condition, shape, or fit. For a grille, accuracy matters beyond overall dimensions: edge profiles, mounting points, openings, surface transitions, and connections with surrounding bodywork all affect fitment.

MetroY Ultra’s high-accuracy scanning captures these details more completely than manual measurement alone. Instead of relying on visual approximation or limited dimensions, restorers and designers can work from the actual geometry of the part. The resulting digital model provides a reliable reference for repair planning, reproduction, modification, or long-term digital archiving.

By accurately digitizing the grille, MetroY Ultra helps make restoration and redesign more controlled, especially for parts that are no longer easy to replace through normal supply channels.

Scanning a Black Part for Custom Fitment

Another representative application came from a modified S197 setup equipped with a 132 mm Whipple throttle body. The project required a custom thermal spacer kit, where accurate understanding of component interfaces was critical.

In performance modification projects, custom parts often need to fit a specific setup rather than a factory configuration. Important geometry includes mounting areas, bolt holes, edge profiles, clearances, and surrounding structures. Without accurate data, repeated adjustments may be needed to achieve proper fitment.

The component had a black surface, which can make 3D scanning more challenging, especially for curved areas, recessed sections, and fitment-critical edges. Yet MetroY Ultra was able to capture usable geometry across these challenging areas, providing a digital reference for spacer design and alignment.

Using scan data from the actual component helps reduce guesswork before modeling or manufacturing. It allows designers to better understand fitment requirements, clearance needs, and key features required for proper installation.

A More Reliable Starting Point for Reverse Engineering

Together, these examples demonstrate three capabilities that matter in real-world reverse engineering: accurate geometry for highly customized fitment, high-accuracy digitization of existing parts, and reliable scanning of challenging black surfaces. Revo Metro helps connect the scanning stage with downstream work by supporting scan capture, review, and initial data preparation. The result is a workflow that starts from the actual geometry of the part, rather than rough measurements or assumptions.

For designers, makers, restorers, and engineers working with custom parts, legacy components, or small-batch projects, MetroY Ultra offers a practical way to bring real-world parts into a digital workflow. Whether the goal is to build around an existing structure, reproduce a hard-to-source component, or capture a challenging surface, the process begins with reliable geometry. MetroY Ultra helps turn that geometry into data that can support the next stage of design and manufacturing.

CERTIFIED METROLOGY

Revopoint MetroY Ultra

Built for demanding reverse engineering, inspection, and design validation workflows, MetroY Ultra captures complex parts with high accuracy using multiple blue-light scanning modes. It is ideal for digitizing mechanical components, automotive parts, and hard-to-scan surfaces for CAD reconstruction and product development.

Price: from ~$1,899
Discount code: REVOYCSCNX

View at Revopoint.

The post Reverse Engineering Without CAD Data: How MetroY Ultra Turns Physical Parts into Digital Designs (Sponsored) appeared first on CNX Software - Embedded Systems News.

Seeed Studio has expanded its XIAO Plus Series introduced in 2025, with the XIAO SAMD21 Plus and XIAO RP2040 Plus USB-C boards adding several GPIOs and integrated Li-ion battery management to the XIAO (SAMD21) and XIAO RP2040, launched in 2020 and 2021, respectively, while keeping the same form factor.

XIAO SAMD21 Plus

Specifications:

• MCU – Microchip SAMD21G18 Arm Cortex-M0+ microcontroller at up to 48MHz with 256KB Flash,32KB SRAM
• USB – 1x USB Type-C port for power and programming
• Expansion I/Os
  • 2x 7-pin “legacy” headers with 11x analog inputs, 11x digital I/Os, 1x DAC, SPI, UART, and I2C
  • Castellated holes and pads with 16x digital I/Os, I2C,and I2S
  • 3.3V I/O voltage (not 5V tolerant)
• Misc
  • Reset and Boot buttons
  • RGB LED
  • Charge LED
• Power Supply
  • 5V via USB-C port
  • Pads for battery
  • On-board PMIC
• Dimensions – 23.5×17.5 mm
• Temperature Range – -40°C ~ 85°C

The upgraded variant of the XIAO SAMD21 features 16 additional SMD GPIOs and integrated Li-ion battery management. The board supports Arduino, PlatformIO, MicroPython, CircuitPython, Zephyr, and more. Check out the wiki for more details.

The XIAO SAMD21 Plus is sold for $5.90 on Seeed Studio, and should soon be offered as an option on the company’s AliExpress and Amazon stores.

XIAO RP2040 Plus

Specifications:

• MCU – Raspberry Pi RP2040 dual-core Cortex-M0+ up to 133 MHz with 264 kB SRAM
• Storage – 2MB SPI flash
• USB – 1x USB Type-C port for power and programming
• Expansion I/Os
  • 2x 7-pin headers with 4x analog inputs, 11x digital I/Os/PWM, SPI, UART, and I2C; 2.54mm pitch
  • Castellated holes and pads with 15x digital I/Os/PWM, 1x I2C 
  • 3.3V I/O voltage (not 5V tolerant)
• Misc
  • Reset and Boot Buttons
  • Charge and User LEDs
  • RGB LED
• Power Supply
  • 5V via USB-C port
  • Pads for battery
  • On-board PMIC
• Dimensions – 21 x 17.8 mm
• Temperature Range – -20°C ~ 70°C

The upgraded variant of XIAO RP2040 features 15 additional SMD GPIOs and integrated Li-ion battery management. The board supports Arduino, PlatformIO, MicroPython, CircuitPython, TinyGo, Rust, Zephyr, and more. You’ll find more details and instructions to get started on the wiki.

Seeed Studio sells the XIAO SAMD21 Plus for $4.90.

The post XIAO SAMD21 Plus and RP2040 Plus USB-C boards get additional GPIO pins, integrated Li-ion battery management appeared first on CNX Software - Embedded Systems News.

After checking out the hardware of the ACEMAGIC Kron Mini K5 with an unboxing and a teardown in the first part of the review, we’ve now had time to test the Intel Core 3 304 Wildcat Lake mini PC in more depth with Windows 11 Pro.

So in this article, we’ll report our experience with the penta-core mini PC, testing features, running system and AI benchmarks, playing 4K and 8K YouTube videos on Firefox, measuring network throughput (2.5GbE and WiFi 6), evaluating thermal performance, and measuring fan noise and power consumption.

Software overview and feature testing

The System > About window confirms we have a 1.5 GHz (base frequency) Intel Core 3 304 computer paired with 12GB of RAM and 477GB of storage, running Windows 11 Pro 25H2 build 26200.9168.

HWiNFO 64 provides additional details about the 15W Intel Core 3 304 penta-core (1P+4E) Wildcat Lake-U processor, the motherboard with 12GB of LPDDR5 SDRAM, and Intel Wildcat Lake integrated graphics with 1 Xe core running at 1000MHz (up to 2300 MHz).

We also ran GPU-Z to get more information about the iGPU, as we usually do when testing mini PCs. However, the Wildcat Lake SoCs are very new, and the utility has not been updated, resulting in the Intel GPU being identified as “FD81” and several “Unknown” fields.

The PL1 and PL2 power limits are set to 22W (PBP) and 35W (MTP), respectively, with PL4 at 60W. That’s a bit different from the values reported for the Beelink EQi Wildcat Lake 304: 15W (PBP), 35W (MTP), and 80W (PL4), so multicore performance should be a little higher by default if the cooling system is sized adequately.

On the memory side, HWiNFO64 confirms that the system comes with 12GB of soldered LPDDR5 SDRAM, consisting of four 3GB Samsung chips.

Task Manager shows that Windows can use up to 11.7GB, running at 5,600 MT/s, while 332 MB is reserved for the hardware, notably the GPU and NPU, both of which are also detected in Task Manager.

Let’s go back to HWiNFO64 to check out networking. The 2.5GbE port relies on a Realtek RTL8125 Gaming 2.5GbE Family Ethernet Controller.

A Realtek RTL8852BE Wi-Fi 6 802.11ax PCIe adapter (module) shows a maximum link speed of 573 Mbps, which should be OK, but is quite lower than the link speed on most other mini PCs. For reference, we had a 2999 Mbps link on the Beelink EQi Wildcat Lake.

We went back to Device Manager to check the Bluetooth version. “LMP13.xxx” firmware version looks up to Bluetooth 5.4.

We also took the opportunity to quickly test Bluetooth connectivity using a file transfer from the mini PC to a Redmi Note 14 Pro 5G smartphone. No issues.

The ACEMAGIC Kron Mini K5 (Intel Core 3 304) has six USB 3.2 Type-A ports and one USB 3.2 Type-C port. Some ports are 5 Gbps, others are 10 Gbps, but sadly none of them are properly marked on the mini PC. Nevertheless, we double-checked the actual speeds using an ORICO M234C3-U4 M.2 NVMe SSD enclosure, HWiNFO 64 utility, and CrystalDiskMark benchmark.

For reference, here are the results for the front left USB 3.2 Type-C (10 Gbps) port… (note: advertised as 20 Gbps)

… and the left USB 3.2 Type-A (5 Gbps) port on the rear panel.

Summary for all seven ports from left to right:

• Front panel
  • USB-C – USB 3.2 – USB 3.2 Gen2 (SuperSpeedPlus+ 10 Gbps) – Read speed: 1,062.51 MB/s; Write speed: 1,039.35 MB/s
  • USB-A #1 – USB 3.2 – USB 3.2 Gen2 (SuperSpeedPlus+ 10 Gbps) – Read speed: 1,060.28 MB/s; Write speed: 1,045.51 MB/s
  • USB-A #2 – USB 3.2 – USB 3.2 Gen2 (SuperSpeedPlus+ 10 Gbps) – Read speed: 1,057.36 MB/s; Write speed: 1,040.01 MB/s
• Rear panel
  • USB-A port #1 (top, left)– USB 3.2 – USB 3.2 Gen1 (SuperSpeed) – Read speed: 455.90 MB/s; Write speed: 463.75 MB/s
  • USB-A port #2 (top, right) – USB 3.2 – USB 3.2 Gen1 (SuperSpeed) – Read speed: 455.89 MB/s; Write speed: 463.55 MB/s
  • USB-A port #3 (bottom, left) – USB 3.2 – USB 3.2 Gen1 (SuperSpeed) – Read speed: 455.98 MB/s; Write speed: 464.15 MB/s
  • USB-A port #4 (bottom, right) – USB 3.2 – USB 3.2 Gen1 (SuperSpeed) – Read speed: 455.97 MB/s; Write speed: 464.51 MB/s

Since nothing is marked on the mini PC, you’ll just have to remember that the faster ports are all located on the front panel. Note that the USB-C port is advertised as a “USB 3.2 Gen2×2 port up to 20 Gbps”, but we couldn’t test this with our peripherals.

The Kron Mini K5 supports up to three independent displays via HDMI 2.0, DisplayPort 1.4, and USB-C with DP 1.4 Alt Mode.

We verified this by connecting three displays to the mini PC: a KTC A32Q8 32-inch 4K Smart Monitor via DisplayPort 1.4, a CrowView 14-inch laptop monitor via HDMI 2.0, and a CrowVi 15.6-inch display via USB-C. All three screens worked fine in extended desktop mode.

ACEMAGIC Kron Mini K5 (Intel Core 3 304) benchmarks on Windows 11 Pro

Before running benchmarks on the mini PC, we set the Power Mode to Best Performance. Note that the tests were performed in a room with an ambient temperature of about 28–29°C, so your own results may differ depending on room temperature and system settings.

The Kron Mini K5 mini PC scored 5,309 points in PCMark 10, broken down into Essentials (8,266 points), Productivity (9,758 points), and Digital Content Creation (5,035 points).

3DMark Fire Strike is a CPU+GPU workload. The system scored 1,880 points.

The system achieved an overall score of 2,686.2 points in PassMark PerformanceTest 11.

The Disk Mark score is a little lower than a typical score in a mini PC with an NVMe SSD. CrystalDiskMark was used to confirm the mid-range performance of the included 512GB NVMe SSD, which reached 1,780.49 MB/s for sequential reads and 1,675.72 MB/s for sequential writes.

Cinebench R23 is useful to evaluate both single-core and multi-core CPU performance and get a first idea of the mini PC’s cooling ability.

The Kron K5 mini PC scored 1,729 points in the single-core test and 5,027 points in the multi-core benchmark, with an MP ratio of 2.91x on a heterogeneous (1P+4E) penta-core processor.

Let’s test the GPU with Unigine Heaven Benchmark 4.0. The system averaged 18.8 FPS and scored 473 points at the standard 1920×1080 resolution, confirming that the Intel Xe3 Graphics in the Core 3 304 is better suited to general graphics work and everyday use than demanding 3D gaming.

Next up is 4K and 8K YouTube playback in Firefox, tested at 30 and 60 FPS.

The results showed that the Kron Mini K5 could play YouTube videos from 4K 30 FPS through 8K 60 FPS smoothly, with no obvious stuttering and no (or few) dropped frames. That’s the same result as on the Beelink EQi Wildcat Lake mini PC, confirming the Intel Core Series 3 processors are solid choices for video playback.

The Intel Core 3 304 SoC comes with Intel Xe3 Graphics delivering up to 9 TOPS of AI performance, and an Intel AI Boost NPU delivering up to 15 TOPS. We therefore tested performance with Geekbench AI using OpenVINO on the GPU and NPU.

On the GPU, the Kron K5 mini PC achieved 2,752 points for the single-precision score, 9,722 points for the half-precision score, and 11,787 points for the quantized score. The full results are available on the Geekbench website.

On the NPU, the scores were: 850 points for single precision, 16,657 points for half precision, and 26,212 points for quantized. The full results are available on the Geekbench website. The results show that the NPU has a clear advantage with quantized models.

ACEMAGIC Kron Mini K5 Windows 11 benchmarks comparison against other mini PCs

Now that we have benchmark data, let’s compare the Windows 11 benchmark results of the ACEMAGIC Kron mini K5 against other entry-level to mid-range mini PCs: Beelink EQi Wildcat Lake and GEEKOM A5 Pro 2026 Edition (AMD Ryzen 5 7530U).

Here are the basic specifications for the three systems

Let’s compare the benchmark results

As one would expect, the two Intel Core 3 304 mini PCs have similar scores, but there are still some interesting differences.  The slower 5600 MT/s LPDDR5 memory on the Kron K5 leads to a lower Memory Mark score and appears to somewhat negatively impact 3D graphics and AI scores compared to the Beelink EQi Wildcat Lake. While the NVMe SSD on the ACEMAGIC mini PC is fairly low-end, it’s still faster than the UFS 3.1 storage on the Beelink device.

The GEEKOM A5 Pro 2026 Edition mini PC performs better with its AMD Ryzen 5 7530U 6-core/12-thread processor, especially on multi-core and 3D graphics workloads. The NVMe SSD is also significantly faster. It does cost $100+ more than the Wildcat Lake systems.

Network performance testing (2.5GbE and Wi-Fi 6)

We tested the 2.5GbE port of the Kron Mini K5 (Intel Core 3 304) mini PC using the iperf 3.21 utility and an UP Xtreme i11 Edge mini PC (192.168.31.12) on the other end of the connection.

• Download

PS C:\Users\aey\Downloads\iperf3.21_64\iperf3.21> .\iperf3.exe -t 60 -c 192.168.31.12 -i 10 -R
Connecting to host 192.168.31.12, port 5201
Reverse mode, remote host 192.168.31.12 is sending
[  5] local 192.168.31.116 port 61912 connected to 192.168.31.12 port 5201
[ ID] Interval           Transfer     Bitrate
[  5]   0.00-10.00  sec  2.76 GBytes  2.37 Gbits/sec
[  5]  10.00-20.00  sec  2.76 GBytes  2.37 Gbits/sec
[  5]  20.00-30.01  sec  2.77 GBytes  2.37 Gbits/sec
[  5]  30.01-40.00  sec  2.76 GBytes  2.37 Gbits/sec
[  5]  40.00-50.01  sec  2.76 GBytes  2.37 Gbits/sec
[  5]  50.01-60.01  sec  2.76 GBytes  2.37 Gbits/sec
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bitrate         Retr
[  5]   0.00-60.01  sec  16.6 GBytes  2.37 Gbits/sec    0            sender
[  5]   0.00-60.01  sec  16.6 GBytes  2.37 Gbits/sec                  receiver

iperf Done.

• Upload

PS C:\Users\aey\Downloads\iperf3.21_64\iperf3.21> .\iperf3.exe -t 60 -c 192.168.31.12 -i 10
Connecting to host 192.168.31.12, port 5201
[  5] local 192.168.31.116 port 62092 connected to 192.168.31.12 port 5201
[ ID] Interval           Transfer     Bitrate
[  5]   0.00-10.01  sec  2.77 GBytes  2.38 Gbits/sec
[  5]  10.01-20.01  sec  2.76 GBytes  2.37 Gbits/sec
[  5]  20.01-30.00  sec  2.76 GBytes  2.37 Gbits/sec
[  5]  30.00-40.01  sec  2.76 GBytes  2.37 Gbits/sec
[  5]  40.01-50.01  sec  2.76 GBytes  2.37 Gbits/sec
[  5]  50.01-60.00  sec  2.76 GBytes  2.37 Gbits/sec
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bitrate
[  5]   0.00-60.00  sec  16.6 GBytes  2.37 Gbits/sec                  sender
[  5]   0.00-60.02  sec  16.6 GBytes  2.37 Gbits/sec                  receiver

iperf Done.

The 2.5GbE port reached 2.37 Gbps in both upload and download. All good.

We added a Xiaomi Mi Router AX6000 to the mix to test Wi-Fi 6 5 GHz.

• Download

PS C:\Users\aey\Downloads\iperf3.21_64\iperf3.21> .\iperf3.exe -t 60 -c 192.168.31.12 -i 10 -R
Connecting to host 192.168.31.12, port 5201
Reverse mode, remote host 192.168.31.12 is sending
[  5] local 192.168.31.128 port 57583 connected to 192.168.31.12 port 5201
[ ID] Interval           Transfer     Bitrate
[  5]   0.00-10.01  sec   983 MBytes   824 Mbits/sec
[  5]  10.01-20.01  sec  1007 MBytes   845 Mbits/sec
[  5]  20.01-30.01  sec  1018 MBytes   854 Mbits/sec
[  5]  30.01-40.01  sec   973 MBytes   816 Mbits/sec
[  5]  40.01-50.00  sec  1023 MBytes   859 Mbits/sec
[  5]  50.00-60.01  sec   987 MBytes   827 Mbits/sec
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bitrate         Retr
[  5]   0.00-60.02  sec  5.85 GBytes   838 Mbits/sec   64            sender
[  5]   0.00-60.01  sec  5.85 GBytes   838 Mbits/sec                  receiver

iperf Done.

• Upload

PS C:\Users\aey\Downloads\iperf3.21_64\iperf3.21> .\iperf3.exe -t 60 -c 192.168.31.12 -i 10
Connecting to host 192.168.31.12, port 5201
[  5] local 192.168.31.128 port 57573 connected to 192.168.31.12 port 5201
[ ID] Interval           Transfer     Bitrate
[  5]   0.00-10.01  sec   962 MBytes   807 Mbits/sec
[  5]  10.01-20.00  sec   979 MBytes   821 Mbits/sec
[  5]  20.00-30.01  sec  1.00 GBytes   860 Mbits/sec
[  5]  30.01-40.00  sec   944 MBytes   792 Mbits/sec
[  5]  40.00-50.02  sec  1006 MBytes   843 Mbits/sec
[  5]  50.02-60.01  sec  1.00 GBytes   860 Mbits/sec
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bitrate
[  5]   0.00-60.01  sec  5.80 GBytes   831 Mbits/sec                  sender
[  5]   0.00-60.03  sec  5.80 GBytes   830 Mbits/sec                  receiver

iperf Done.

Download speed was 838 Mbps and upload speed was 830 Mbps, which is a decent result for Wi-Fi 6 and sufficient for high-resolution video streaming, file transfers, and general network use. The maximum link speed of 573 Mbps reported in HWiNFO 64 appears to be erroneous…

Thermal performance

We ran 3DMark Fire Strike CPU+GPU workload on the mini PC to test it for potential CPU throttling. The maximum CPU temperature was about 90°C, and HWiNFO 64 detected thermal throttling on all Efficient cores, but not on the Performance core. This is different behaviour from the Beelink EQi Wildcat Lake, which didn’t have any thermal throttling, mainly because of the lower PL1 thermal limit (15W vs 22W), and the Beelink mini PC exceeded power limits instead.

After rebooting the system to run Cinebench R23 multi-core, the maximum CPU temperature rose to about 92°C, and thermal throttling was observed on all five cores this time around.

The different thermal strategy doesn’t seem to impact performance that much. However, the 15W configuration should be better for sustained loads. We also ran the test in a room at 28-30°C, so it’s possible that in a cooler room (20°C), the ACEMAGIC Kron Mini K5 wouldn’t have reached thermal limits.

Fan noise

The Kron Mini K5’s fan is not particularly noisy at idle and during light workloads, but under high load the fan becomes noticeably louder. We used a sound level meter placed about 5 cm above the top of the chassis.

• Idle – 37.4 – 38.0 dBA
• 3DMark (Fire Strike) – 38.0 – 41.6 dBA, rising to 41.2–46.0 dBA during heavy scenes
• Cinebench R23 multicore – 45.7 – 47.0 dBA

The test room had a background noise level of about 36.5–37.0 dBA. For reference, the Beelink EQi Wildcat Lake mini PC was measured at just 39.8 – 40.4 dBA when running the Cinebench R23 multicore benchmark. So the Beelink is a better system if you need a relatively quiet device, and it’s one of the quietest actively-cooled mini PCs we’ve ever reviewed.

Kron Mini K5’s power consumption

The Kron Mini K5 mini PC can be powered by either a 19V/3.42A adapter through a DC jack or a USB-C PD adapter.

We used a wall power meter to measure power consumption with the provided 19V power adapter:

• Power off – 0.7 – 0.8 W
• Idle – 3.7 – 4.3 W
• YouTube 8K 60fps in Firefox – 12.5 – 15.9 W
• 3DMark (Fire Strike) – 13.0 – 17.9 W (during combat scenes)
• Cinebench R23 Multi-Core
  • First few seconds: 35.7 – 37.1 W
  • Sustained: 29.7 – 33.0 W

We recently received a Sabrent Voltik 252W 8-port charging station with a built-in per-port power meter function, so we used it to measure the power consumption through USB-C. This is a bit different, since the losses in the charging station are not part of the measurements, and the values are somewhat lower. It’s better if you want to focus on the mini PC’s power consumption per se, but not if the purpose is to evaluate the electricity costs.

USB-C power measurements:

• Powered off – 0.2-0.4 W
• Idle – 2.8 – 3.4 W
• YouTube 8K 60fps in Firefox – 11.6 – 17.5 W
• 3DMark (Fire Strike) – 12.0 – 18.9 W (during combat scenes)
• Cinebench R23 multicore
  • First few seconds: 31.1 – 33.7 W
  • Sustained: 26.7 – 32.7 W

Note that during testing, the mini PC was connected to Wi-Fi 6, a USB-A RF dongle for a wireless keyboard, a USB mouse, and a 32-inch 4K KTC A32Q8 monitor (HDMI 2.0).

Conclusion

The ACEMAGIC Kron Mini K5 mini PC works well with Windows 11 Pro. It delivers strong single-core performance, 2.5GbE and WiFi 6 networking, supports up to three 4K displays through HDMI, DP, and USB-C ports, and plays 4K and 8K YouTube videos smoothly up to 60 FPS. The Intel Core 3 304 is also suitable for light AI workloads, although it doesn’t qualify as a Copilot+ PC with a 9 TOPS GPU and 15 TOPS NPU.

In many ways, it offers similar performance and features as the Beelink EQi Wildcat Lake Core 3 304 mini PC. On the downside, it lacks high-speed interfaces like 10GbE and Thunderbolt 4/USB4, the memory has lower capacity and bandwidth (12GB LPDDR5-5600 vs 16GB LPDDR5-6400), and its fan is not as quiet as the one on the Beelink mini PC. A different power limit configuration (22W vs 15W PL1) means the mini PC hit thermal limits instead of power limits in our stress tests. While the 512 GB NVMe SSD included in the ACEMAGIC Kron Mini K5 can be considered “mid-range”, it’s still faster than the 512 GB UFS found in the Beelink EQi Wildcat Lake. Having said that, both have two NVMe SSD sockets, so upgrades are possible.

In the third part of the review, we will install and test Ubuntu 26.04 on the Kron Mini K5 mini PC to see how Linux performs on the Intel Core 3 304 mini PC.

We’d like to thank ACEMAGIC for sending the Kron Mini K5 (Intel Core 3 304) mini PC for review. It is sold for $399.99 on Amazon, a fair price compared to the $469 asked for the Beelink EQi Wildcat Lake (16GB RAM, 10GbE, and Thunderbolt 4) and the $380 price for the BOSGAME ECO with a similar 12GB/512GB configuration, but a different port layout.

CNXSoft: This article is a translation – with some additional insights – of the original review on CNX Software Thailand by Suthinee Kerdkaew.

The post ACEMAGIC Kron Mini K5 Review – Part 2: A $400 Intel Core 3 304 Wildcat Lake mini PC tested with Windows 11 Pro appeared first on CNX Software - Embedded Systems News.

OpenTrailPaper is an open-source, DIY e-paper bike computer project comprised of firmware for the ESP32-S3-based T5 E-Paper S3 Pro devkit and Android and iOS companion apps for maps, routes, and settings.

It shows ride data and offline maps, follows standard GPX (GPS Exchange Format) routes, records FIT (Flexible and Interoperable Data Transfer used in Garmin devices) files, and connects to Bluetooth (and soon ANT/ANT+) heart rate, power, and cadence sensors.

Before looking into the project’s details, let’s check out the LILYGO T5 E-Paper S3 Pro specifications:

• ESP32-S3-WROOM-1 wireless module
  • SoC –  ESP32-S3N16R8 dual-core Tensilica LX7 microcontroller @ up to 240 MHz with
    • 2.4 GHz 802.11n WiFi 4 and Bluetooth 5.0 LE connectivity
    • Memory – 8MB PSRAM
    • Storage – 16MB SPI flash
  • PCB antenna
• Storage – MicroSD card slot
• Display – 4.7-inch e-Paper display with 960×540 resolution, 16 shades of gray, capacitive touch (GT911); ED047TC1 drive IC
• Wireless
  • 2.4 GHz 802.11n WiFi 4 and Bluetooth 5.0 LE connectivity (ESP32-S3)
  • LoRa module (SX1262) supporting 433 MHz, 868 MHz, and 915 MGz bands
  • GNSS – u-blox MIA-M10Q or Quectel L76K module (autodetected by the firmware)
• USB – USB Type-C port for power and programming
• Expansion – 4-pin Qwicc connector
• Misc
  • Power, Reset, Boot, and User (IO48) buttons
  • PCF8563 RTC
• Power Supply
  • 5V via USB Type-C port
  • 3.7V, 1,500mAh battery good for under 8 hours per charge with OpenTrailPaper firmware and front light off; (note: might be optimized)
  • BQ25896 charger and BQ27220 fuel gauge
• Dimensions – 129 x 69 x 11 mm

The LILYGO T5 ships with either the company’s factory firmware or MeshCore for off-grid communication and location sharing. If you’re going to flash OpenTrailPaper on the device, simply purchase the LILYGO firmware version since you’ll save a few dollars that way. You’ll find additional technical details and code samples for low-level hardware features of the T5 E-Paper S3 Pro on GitHub.

Back to the OpenTrailPaper e-paper bike computer project specifically. The project’s GitHub repository features the source code for the ESP32-S3 firmware based on the ESP-IDF framework, the OpenTrailPaper Android and iOS companion apps, as well as tools to create compatible maps from a smartphone using the companion app, a Python script in a terminal, or JavaScript code running in a web browser. Everything has been developed with Anthropic Claude as an assistant, but with human supervision, and it’s a working project, albeit with some caveats for now (more on that below). Further documentation can also be found on the project’s website.

The easiest way to get started is to go to the setup page in a browser supporting the Web Serial API. You’ll go through the instructions to flash the firmware over USB and build offline map tiles stored on a microSD card.

The companion app can plan and transfer routes, build offline maps, copy recorded rides, change settings, and install firmware updates over Bluetooth. The bike computer itself does not need the phone during a ride.

Here’s what runs on the bike computer (during a ride) and the smartphone (before or after a ride).

One of the main reasons for getting an e-paper bike computer is to be able to clearly see data and the map while riding outdoors. I don’t usually need GPS when I ride a bicycle where I live, but when I travel, I typically connect my smartphone running the MAPS.me offline maps app to a bicycle stand to get directions. In most cases, it’s not super convenient to use since it’s unreadable unless I find/make some shadows.

To reproduce the project, you’ll need to purchase the LILYGO T5 E-Paper S3 Pro for $90 on Amazon or $103 on AliExpress, as well as find a way to mount it to your bicycle. A phone bicycle stand should probably do. Bear in mind that there are a few limitations:

• The T5 S3 Pro is not waterproof, so you may have to put it in a ZIP bag to protect it from rain.
• No pressure sensor. Climbing is derived/estimated from elevation data in the map tiles
• GPS could be improved – A multi-band GPS module would lock on faster and hold a fix better in various conditions
• No magnetometer – The map only orients using direction of travel, so it doesn’t work while you’re stopped.
• Impractical, small hardware buttons and the touch panel might not work well with winter gloves or in the rain.

The post OpenTrailPaper transforms LILYGO T5 E-Paper S3 Pro devkit into a DIY e‑paper bike computer appeared first on CNX Software - Embedded Systems News.

Arduino has just announced the release of ArduinoCore-Zephyr version 1.0.0, with new hardware support, including the VENTUNO Q SBC, an updated Zephyr 4.4.1 base, support for the Nicla Vision camera, and various fixes and improvements.

Arduino first announced its plans to switch from Arm Mbed to Zephyr RTOS in July 2024, before releasing the first beta of Arduino Core for Zephyr in December of the same year. ArduinoCore-Zephyr 1.0.0 is not the first stable version, since that was version 0.90.0 released on July 29, 2026. The new release builds on that, so let’s see what’s new.

ArduinoCore-Zephyr 1.0.0 highlights

• Support for Arduino VENTUNO Q SBC, specifically the STM32H5F5 real-time MCU on the board
• Update to Zephyr 4.4.1 with the latest upstream improvements, security fixes, and driver support
• Added Nicla Vision camera support to capture and process image data directly through the Zephyr core
• Various bug fixes and CI improvements
• Currently supported platforms from Arduino
  • UNO Q board
  • VENTUNO Q board
  • Portenta H7 board
  • Portenta C33 board
  • Opta microPLC
  • GIGA R1 WiFi board
  • Nano Matter board
  • Nano RP2040 Connect board
  • Nano 33 BLE board
  • Nicla Sense ME module
  • Nicla Vision module

You’ll find the full list of changes on the project’s GitHub release page.

If you want to check it out, download and install the latest version of the IDE, update the board files if requested, and install or update Arduino Zephyr Boards to 1.0.0. If you want to test the latest release on the Renesas EK-RA8D1, NXP FRDM-MCXN947, or NXP FRDM-RW612 board, install Zephyr Community Boards, also maintained by Arduino.

That’s about all you have to do in the IDE before building the sketch. The Arduino IDE should automatically upload and install the Zephyr loader when uploading the firmware for the first time, but in some cases, you may need to manually install the Zephyr loader onto your board. A manual update appears to be mandatory for the Portenta C33.

The post ArduinoCore-Zephyr 1.0.0 adds support for Arduino VENTUNO Q, updates to Zephyr 4.4.1, and more appeared first on CNX Software - Embedded Systems News.

If you are not into ducks, but still would like to play around with physical AI, the Petoi Quaddle is a mini robot dog equipped with 4 servos controlling its legs and is designed to help people learn robotics, coding, AI, and 3D printing.

The robot is powered by an ESP32-S3 motion core for servo control and wireless connectivity, and an optional ESP32-S3 AI core to enable voice commands and Smart Home control. The robot can walk, strafe, spin, talk back, and react to the user’s touch.

Three models are available depending on the user’s requirements: Builder, Buddy, and Scout.

A Raspberry Pi Zero can also be mounted on the top in place of the ESP32-S3-based AI core. Petoi Quaddle runs OpenCat open-source firmware (Arduino) already used in the earlier Petoi Bittle robot dog, first introduced in 2020. The robot can be programmed in a visual programming IDE with drag-and-drop blocks, as well as Python, MicroPython, Arduino, C++, and ROS. A Raspberry Pi Zero (UART) or a computer (Wi-Fi or Bluetooth) is needed for the latter.

Other highlights include compatibility with LEGO and 3D-printed shells, partnerships with TuyaOpen for Smart Home integration, the ability to program AI Code to use ESPHome or Xiaozhi AI, mobile apps for control, and more. The documentation is already up but under construction, and you can already play around with the Quaddle simulator.

Petoi has launched the Quaddle on Kickstarter with a $50,000 funding target. The three main rewards are as follows:

• Quaddle Builder – $99 ($149 MSRP)
• Quaddle Buddy – $139
• Quaddle Scout – $199

You can order it as a kit or preassembled. Shipping adds $15 to most of the world and $18 to the European Union. The company expects to ship rewards in December 2026.

The post Petoi Quaddle – A mini robot dog for physical AI experimentation (Crowdfunding) appeared first on CNX Software - Embedded Systems News.

CNXSoft: This is a guest post by Pavel Burgr, who recently added Puya PY32 MCU support to TinyGo 0.42, the latest version of the Go compiler for microcontrollers. We first covered TinyGo in 2019, and the project is still going strong with 17.7k stars, 1.1k forks, and 243 contributors. I just made a few edits and added illustrations and a few links to the article.

TinyGo 0.42 adds support for the Puya PY32 family of 32-bit Arm microcontrollers. The new port includes startup code, generated device-register definitions, linker scripts, GPIO and UART support, and targets for 87 individual MCU configurations. It also includes board definitions for two inexpensive EmbedFire development boards available from AliExpress.

The port was contributed by Pavel Burgr and merged through TinyGo pull request #5106. One motivation was to establish a solid, easy-to-use firmware platform for small IoT devices built around the Registry service and the BleRiot low-power radio protocol (more on that at the end of the post). Those projects are separate from TinyGo, but they show the sort of resource-constrained application the new targets are intended to enable.

Broad PY32 target coverage

TinyGo is a Go compiler for microcontrollers, WebAssembly, and other constrained systems. It uses LLVM to produce compact native binaries and provides a hardware abstraction layer through packages such as machine, while retaining familiar Go syntax and tooling. Running Go on a part with only 2 KB of SRAM may sound a little like a dare, but this is precisely the sort of dare TinyGo was made for.

The PY32 port covers both Cortex-M0+ and Cortex-M4 devices. The target table contains 87 concrete memory configurations across 40 device-definition families, ranging from parts with 8 KB of flash and 2 KB of SRAM to devices with 512 KB of flash and 144 KB of SRAM.

Supported series include:

• PY32E407
• PY32F001, F002, F003, F021, F030, F031, F032, F040, F071, F072, F403, and F410
• PY32L020 and L090
• PY32M010, M020, M030, M031, and M070
• PY32MD310, MD320, MD410, MD420, and MD430
• PY32T020, T090, and T092

Each chip target selects the correct flash and SRAM map. Peripheral definitions are generated from Puya’s CMSIS data through the supporting py32-svd repository, giving driver authors typed register names in device/py32 instead of private address tables.

The machine API in TinyGo 0.42 covers clock and timer startup, GPIO, and UART, including a second UART on selected parts. ADC, I2C, SPI, and PWM do not yet have portable machine drivers, but their generated registers are available to low-level code.

Two supported EmbedFire boards

Two small Puya32 development boards have dedicated targets and pin aliases, making them the simplest way to try the port. Both are available for under $2 on AliExpress, and the PY32F030 board goes for $7 and up on Amazon.

The PY32F030 board maps LED1, LED2, and LED3 to PA2, PA3, and PA4, and its two keys to PA5 and PA6. Its default UART uses PA7 for TX and PA8 for RX. The smaller PY32F002B board maps the LEDs to PA1, PA5, and PA4, the keys to PA3 and PA0, and the default UART to PA6 and PA7.

Both boards define machine.LED as an alias for the second onboard LED. As embedded tradition demands, the same first program can perform the ceremonial blink on either target:

package main

import (
	"machine"
	"time"
)

func main() {
	led := machine.LED
	led.Configure(machine.PinConfig{Mode: machine.PinOutput})

	for {
		led.Set(true)
		time.Sleep(500 * time.Millisecond)
		led.Set(false)
		time.Sleep(500 * time.Millisecond)
	}
}

With a supported SWD probe connected and pyOCD configured for the selected Puya device, the program can be built and flashed with one command, with no Makefile archaeology required:

tinygo flash -target=embedfire-py32f030 .

For the smaller board, only the target changes:

tinygo flash -target=embedfire-py32f002b .

The board targets inherit the exact py32f030x8 and py32f002bx5 memory maps and invoke the matching pyOCD target during flashing. Bare-chip targets are also available for custom boards. A small TinyGo PY32 project template provides additional blink examples and a starting Makefile.

From low-cost MCU to Registry-connected IoT node

With the traditional blink out of the way, the practical motivation behind the port is more interesting. Pavel developed the PY32 support while building BleRiot, a simple and reliable request/response protocol for low-power RF nodes.

Each node exposes signed 32-bit registers and understands just two operations: GET and SET. That is the whole vocabulary, which is a feature rather than an omission. The Linux hub polls registers and handles names and unit conversions as well as retries and timeouts. This small, predictable interface keeps firmware compact, while recovery from a dropped packet is handled once in the hub instead of by every application.

One hub communicates with many nodes over a custom 250 kbps GFSK radio link with BLE-compatible packet framing, although it is not Bluetooth LE or GATT. The reference node pairs a PY32F030x8 with a PAN211x 2.4 GHz radio, while the hub uses a simple USB radio dongle.

BleRiot publishes the node values to Registry, a lightweight local service that gives each register a name, current value, optional metadata, and a time-to-live. Stale values expire automatically. Applications can use its HTTP/JSON API directly; the supplied Node-RED nodes combine with standard MQTT nodes for bridging, and a built-in Prometheus endpoint exposes numeric registers as gauges ready for Grafana dashboards. In other words, getting a temperature onto a dashboard does not require teaching a 3 KB MCU about Grafana.

A controller follows the same path in reverse when it requests a change. BleRiot shares device types and register definitions between the TinyGo firmware and Linux hub, and its tooling can generate a node identity, build the firmware with TinyGo, and flash it over SWD.

Registry and BleRiot are not required to use TinyGo on PY32, and the PY32 support is not tied to a particular radio. They are useful examples of why low-cost microcontrollers benefit from a maintained compiler, reproducible chip targets, and a standard Go-facing hardware API.

Availability and next steps

PY32 support is available in TinyGo 0.42.0. The two EmbedFire targets provide the most direct entry point, while the broader chip list gives developers a foundation for custom boards ranging from tiny sensor nodes to larger Cortex-M4 controllers.

TinyGo 0.42 emphasizes runtime stability, correct memory maps, GPIO, and serial communication rather than claiming complete peripheral coverage. Future contributions can add portable ADC, I2C, SPI, PWM, and board definitions on top of the generated device/py32 register layer. For developers already using Go on the host side of an IoT system, the port makes Puya’s inexpensive MCU family a much more approachable endpoint.

CNXSoft note: TinyGo support for AT32 MCUs is apparently next.

The post TinyGo 0.42 adds support for 87 Puya PY32 MCU variants and two EmbedFire boards appeared first on CNX Software - Embedded Systems News.

Boardcon CM311Y3 system-on-module (SoM) is powered by an Amlogic A311Y3 octa-core Cortex-A78/A55 SoC with an 8 TOPS NPU and targets 4K video surveillance systems, AI edge computing devices, interactive terminals, enterprise thin clients, and autonomous robots.

The module comes with up to 16GB LPDDR5 and up to 256 GB eMMC flash, exposes 210 pins through castellated edges, and a development board is also provided for evaluation and early software development. We don’t often see new SoMs or SBCs based on Amlogic SoCs these days, so let’s have a closer look.

Boardcon CM311Y3 System-on-Module

CM311Y3 specifications:

• SoC – Amlogic A311Y3
  • CPU
    • 2x ARM Cortex-A78 cores @ 2.4GHx
    • 6x ARM Cortex-A55 cores @ 2.0GHz
    • RISC-V core for system control processing
    • RISC-V core for Always-on power management and Sensor Hub
  • GPU – Arm Mali-G625 MC1 with support for OpenGL ES 3.2, Vulkan 1.4, and OpenCL 3.0
  • VPU
    • Video Decoder – H.264 up to 4Kp60 H.265, AV1, VP9, AVS3 up to 4Kp120
    • Video Encoder – H.264/H.265 up to 4Kp60
  • ISP – 16MP resolution
  • NPU – 8 TOPS AI accelerator; supports INT4, INT8, INT16, FP8, FP16, and BF16
  • DSP – HiFi 5
• Memory – 4GB, 8GB, or 16GB LPDDR5/LPDDR5X
• Storage – 32GB, 64GB, 128GB, or 256GB eMMC flash
• Networking – Realtek RTL8211H Gigabit Ethernet PHY
• Castellated edges (210 pins) for carrier board
  • Storage – SD card interface
  • Display Interfaces
    • HDMI Tx
    • MIPI DSI/LVDS(A)
    • eDP/MIPI DSI/LVDS/VX1(B)
  • Camera/Video Input – 2x MIPI CSI, HDMI Rx
  • Audio – S/PDIF output
  • Networking – 2x Ethernet (2.5GbE and GbE from A311Y3 specs, but the carrier board below uses 2x GbE)
  • USB – USB 3.0, USB 3.0 DRD, USB 2.0
  • PCIe – PCIe 3.0 x2
  • Low-speed I/Os – UART, 2x I2C, GPIO, PWM, etc…
  • Misc – SDIO, IR Rx, Debug
• Supply Voltage – 3.4 to 5.5V DC
• Dimensions – 52 x 45 mm (8-layer PCB, 0.9 mm pitch QFN)
• Weight – 10.7 grams

The company provides support for Android 16 with Linux 6.12.69 and U-boot 2025.01, and all necessary drivers.  They also offer a “Ubuntu 22.04 cross-compiler environment”, which probably means a VM image with Ubuntu 22.04, development tools, and the SDK pre-installed.

Idea311Y3 Amlogic A311Y3 development board

Idea311Y3 specifications:

• SoM – Boardcon CM311Y3 module described above soldered on board
• Storage
  • MicroSD card slot
  • NVMe SSD via M.2 M-Key slot
• Display Interfaces
  • HDMI 2.1, up to 4K @ 60Hz
  • 2x MIPI DSI, up to 2560 x 1600 @ 60Hz
• Camera/Video Input
  • HDMI 2.1a input, up to 4K @ 60Hz
  • 2x 4-lane MIPI-CSI
• Audio
  • 3.5mm audio I/O jack
  • 2x speakers via 12-pin pluggable terminal block
  • 8-channel audio via HDMI
• Networking
  • 2x Gigabit Ethernet RJ45 ports via Realtek RTL8211F-CG controllers (one on module, one on carrier board)
  • WiFi 5 (802.11ac/a/b/g/n) and Bluetooth 5.4
  • Optional 4G LTE via mini PCIe socket and nano SIM card slot
• USB – 2x USB 3.0 Type-A ports
• Serial port
  • 3-pin serial port for debug
  • RS232 via 12-pin pluggable terminal block
  • RS485 via 12-pin pluggable terminal block
• Misc
  • Reset, Power_EN, and Power buttons
  • RTC with battery connector
  • 2-pin header for USB Boot
• Power Supply – 12V/3A DC input via power barrel jack
• Dimensions – 155 x 115 mm

The development board is supported by the same SDK as the module, and the company can also provide various accessories, namely an EC20 mini PCIe LTE module, an IMX415 8.29 MP CMOS camera module, an IMX307 2.13MP CMOS camera module, a 12V/3A power adapter (EU/US),  and a 10.1-inch 1280 x 800 display (MIPI DSI or LVDS).

Boardcon hasn’t provided any pricing information for the new Amlogic A311Y3 system-on-module and development board. Additional information may be found on the product page.

The post Amlogic A311Y3 Cortex-A78/A55 Edge AI system-on-module delivers up to 8 TOPS appeared first on CNX Software - Embedded Systems News.

Artery Technology AT32F406 and AT32F408 are two new Arm Cortex-M4F MCUs running at up to 216 MHz, with the F408 adding USB OTG High-Speed and an integrated PHY for industrial automation, IoT, USB accessories, and gaming peripherals.

The AT32F406/F408 series both offer DSP instructions and a floating-point unit (FPU), integrate up to 512 KB Flash and 192 KB SRAM, and a 4 KB OTP memory for critical data and parameters. The microcontrollers also offer a QSPI interface for external Flash or RAM expansion, and a range of digital and analog peripheral interfaces.

Artery AT32F406/AT32F408 specifications:

• MCU – 32-bit Arm Cortex-M4F CPU up to 216 MHz; also features Memory Protection Unit (MPU) and DSP instructions
• Memory/Storage
  • 128 KB to 192 KB SRAM
  • 256 KB to 512 KB Flash memory
  • 26 KB boot memory (configurable as a bootloader or general instruction/data memory)
  • 4 KB OTP (one-time programmable) memory
  • QSPI interface for external SPI Flash or SPI RAM with address mapping support
• Peripherals
  • Up to 57 fast GPIOs (AT32F406), 54 GPIOs (AT32F408); almost all 5V-tolerant
  • USB – USB 2.0 OTG High-Speed (OTGHS) controller with on-chip PHY and a dedicated 4 KB buffer (AT32F408 only).
  • Low-speed I/Os
    • Up to 3x I2C interfaces (SMBus/PMBus support)
    • Up to 8x SPI interfaces (up to 36 Mbit/s) with multiplexed half-duplex I2S
    • Up to 6x USART and 2x UART interfaces (supporting LIN, IrDA, RS485 drive enable, and ISO7816)
  • Analog
    • 3x 12-bit 5.33 MSPS A/D converters with up to 16 external input channels, configurable resolution (12/10/8/6-bit), and hardware oversampling up to 16-bit equivalent
    • Dedicated VREF+ pin (-7R type LQFP64 package only)
    • Internal temperature sensor and internal reference voltage
    • EMUXC – External Multiplexer Controller (Enables fast ADC scanning without CPU intervention)
  • Timers
    • 16-bit 8-channel advanced timer with complementary channels for PWM and dead-time generator
    • Up to 7x 16-bit and 1x 32-bit general-purpose timers.
    • 2x 16-bit basic timers.
    • 2x watchdog timers (general and windowed) and 1x 24-bit SysTick timer
    • Enhanced RTC (ERTC) with auto-wakeup, subsecond accuracy, and hardware calendar
  • Clock – 48 MHz High-speed Internal Clock (HICK)
• Misc
  • Infrared transmitter (IRTMR)
  • 2x 7-channel DMA controllers
  • CRC calculation unit
  • 96-bit unique ID
  • SWD and SWO debug interfaces
  • Power-on reset (POR), low voltage reset (LVR), and power voltage monitor (PVM)
• Power – 2.4 V to 3.6 V
• Packages – LQFP64 (7×7 mm); QFN48 (6×6 mm)
• Temperature Range – -40°C to +105°C

On the software side, the AT32F406/F408 are supported by Artery’s AT32 IDE, VS Code extension, Workbench, AI Studio, and AT-Link tools. They also include sLib, which protects selected Flash memory areas from being read while still allowing code execution, helping protect proprietary software. Artery also provides BSP libraries, documentation, and an AT32F405-to-F408 migration guide.

To help developers get started quickly, Artery has released the AT-START-F406 and AT-START-F408 evaluation boards alongside the new MCUs. They feature Arduino-compatible headers to stack existing shields and prototype quickly. At the time of writing, I could only find documentation for both development boards, but purchasing information is not available.

Artery is not exactly well known, but we did write about them in 2023, as a “Black Pill” clone board featured the company’s AT32F403ACGU7 Cortex-M4F MCU, itself a drop-in replacement (but not a direct clone) of ST’s STM32F103 series. The new AT32F406/AT32F408 parts don’t seem to offer similar compatibility against any STM32 SKUs.

The AT32F406 and AT32F408 MCUs are currently in sampling, with mass production underway. More details and documentation are available on the respective product pages and in the press release.

The post Artery AT32F406 and AT32F408 Cortex-M4F MCUs feature optional high-speed USB OTG PHY appeared first on CNX Software - Embedded Systems News.

Layered Ventures LLC has launched the TickrCast, an expandable HUB75 LED ticker display built around an ESP32-S3. The TickrCast ships as a full consumer product designed to display live esports data, traditional sports scores, social media metrics, and flight tracking across a daisy-chained array of panels.

Unlike a typical LED matrix development board, the system is a complete product with a custom controller board, plug-and-play panels, power connectors, a companion app, and an expandable enclosure. The display starts with two panels and can be expanded by adding additional panels, with the company targeting configurations of 10 or more panels. The system is controlled through Wi-Fi, while Bluetooth LE is used for initial provisioning, and the manufacturer provides more than 30 plugins at launch with no subscription required. The platform also supports signed over-the-air (OTA) firmware updates and hardware flash encryption for improved device security.

TickrCast controller specifications:

• ESP32-S3-WROOM-1U wireless module
  • SoC –  ESP32-S3
    • CPU – Dual-core Tensilica LX7 microcontroller @ up to 240 MHz
    • Memory – 8MB PSRAM
    • Storage – 16MB SPI flash
    • Wireless – 2.4 GHz 802.11n WiFi 4 and Bluetooth 5.0 LE connectivity
  • IPEX antenna connector
• Display – HUB75 based LED matrix panels, 64×32 resolution per panel
• Connectivity – 2.4 GHz Wi-Fi and Bluetooth LE (Via ESP32-S3 module)
• USB – USB Type-C port located on the left edge for power and data/programming
• Expansion – Modular design starting at 2 panels, expandable to 10+ panels (panels bolt together for easy servicing)
• Misc – EN (Reset) and BOOT buttons
• Power – High-power 5V-15A JST box-type connector for power delivery
• Dimensions – TBD

The TickrCast runs custom ESP32-S3 firmware for HUB75 matrices, with a web flasher, web dashboard, and mobile apps (coming soon to the App Store & Google Play) for setup and control, along with signed OTA updates, flash encryption, and a plugin system for live esports, sports, social, music, news, weather, flights, and custom graphics. The company says the platform is already running on a 24/7 soak-test fleet, with Bluetooth provisioning, local content caching, a live plugin marketplace, and 60+ official plugins, while developers can publish their own plugins and use an MCP hook to generate them from plain-English requests.

The creator, Joseph Capehart, mentions that handling Wi-Fi and HUB75 Direct Memory Access (DMA) simultaneously on the ESP32-S3 caused conflicts during development, eventually requiring a platform migration. TickrCast solves this by moving the content composition to a server-side rendering and segment caching system, leaving the ESP32-S3 to handle the display output. This helps keep scrolling smooth and updates responsive, while the cloud infrastructure has been load-tested with up to 22,000 concurrent devices on a single server. Note that the project is not open-source, which may be an issue for a device connected to the cloud. However, the developer promises to release the server and firmware code as open source if the company ever shuts down.

Previously, we have written about similar projects that put live data on an LED display, including Meterbit’s Pixlpal, an open-source ESP32-S3 128×64 RGB LED matrix ticker for news, crypto, and smart-home alerts. We have also written about Pimoroni’s Interstate 75 W (RP2350) and Adafruit’s PioMatte; both are designed to control those same HUB75 matrices.

The TickrCast is live on Kickstarter, where a single-panel foundational unit starts at $99 for early birds, while the standard 2-panel “Enthusiast” kit is priced at $159. For those wanting maximum screen real estate, an 8-panel kit goes for $699, and a 10-panel kit for $799. Deliveries are scheduled to begin in February 2027. Stretch goals include injection-molded enclosures at $20,000, native Home Assistant integration at $175,000, and a free extra panel for every backer at $500,000. The company also mentions that the first 50 units will ship with a 3D-printed enclosure.

Note: The company says that, because a grounding issue was discovered in testing, the 10-panel hardware was respun. Backers of this tier will receive 8 panels initially, with the final 2 shipping later once the revised board passes validation.

The post TickrCast – An ESP32-S3 HUB75 LED ticker display with server-side rendering and OTA (Crowdfunding) appeared first on CNX Software - Embedded Systems News.

Harald Kreuzer has built an ESP32-P4 weather station and air quality monitor that features a 10.1-inch dashboard display and a built-in Sensirion SEN66 sensor for particulate matter, CO2, VOC, and NOX, and connects to wireless LoRa and ESP-NOW sensors.

The project is entirely made of off-the-shelf boards, so no custom hardware is used, and it should be relatively easy to reproduce. It doesn’t require any cloud or user account, although it can connect to a choice of weather APIs (Open-Meteo, OpenWeatherMap or Visual Crossing) to display weather forecasts. The build is fully documented in a long blog post, so we’ll highlight the key features of the project in this article.

Hardware specifications for the main unit:

• Based on ESP32-P4-Module-DEV-KIT-C
  • Credit card-sized ESP32-P4-Module-DEV-KIT SBC based on an ESP32-P4NRW32 module with ESP32-P4 with 32MB PSRAM and ESP32-C6 for WiFi and BLE
  • 10.1-inch IPS panel with 1280 x 800 resolution, 800:1 contrast ratio, and 400 cd/m² brightness
  • 8Ω 2W speaker (not used in this project)
  • RPi Camera (B) and 15-pin FFC cable for the camera (not used in this project)
• LoRa connectivity – Heltec V4 (board only)
• Sensors
  • Sensirion SEN66 multisensor for PM1.0, PM2.5, PM4.0, PM10, nitrogen oxides (NOx), and volatile organic compounds (VOCs).
  • Brightness control through
    • BH1750 light sensor
    • DFRobot C4001 24GHz mmWave sensor for human presence detection (12-meter range variant)

Harald is also working on a version based on the 10-inch version of the GUITION ESP32-P4 touchscreen displays, which should probably be thinner since the ESP32-P4 is directly integrated into the display board.

While you could use the main unit as is, you may want to add a few extra LoRa and ESP-NOW modules:

• Sensor-Receiver – LoRa + ESP-NOW receiver based on Heltec WiFi LoRa 32 (ESP32-S3)
• Sensor-LORA-BME280 – Temperature, humidity, and pressure LoRa node with Heltec WiFi LoRa 32 V3/V4
• Sensor-LORA-BME280-HTCC-AB01 – Temperature, humidity, and pressure LoRa node with Heltec CubeCell HTCC-AB01 V2 (cheaper, no display)
• Sensor-LORA-SHT45 – Temperature and humidity LoRa node with Heltec WiFi LoRa 32 (ESP32-S3)
• Sensor-LORA-RadiationD – LoRa Geiger counter based on RadiationD-v1.1 tube and Heltec WiFi LoRa 32 V4 (ESP32-S3)
• Sensor-LORA-RadiationD-HTCC-AB01 – Same as above, but with Heltec CubeCell HTCC-AB01 V2
• Sensor-ESP-NOW-BME280 – Temperature, humidity, and pressure ESP-Node node using a generic ESP32 board (oldest sub-project)

That’s about it for the hardware. The source code for the main unit, the receiver module, and sensor modules can be found on GitHub. It mostly relies on the ESP-IDF framework, except for the CubeCell-based modules whose firmware is based on Arduino/PlatformIO. The main unit renders the user interface with LVGL and was designed with EEZ Studio open-source GUI editor.

You’ll also need to print the enclosures for the main unit and sensors for your 3D printer. The only issue is that the STL files don’t seem to have been released just yet, as I could only find files for an older version of the project, and they won’t be suitable for the new variant. The STEP and SolidWorks files can now be found in the SolidWorks folder of the aforelinked firmware repository.

There are also plenty of DIY weather station and air quality monitor projects such as MoreSense MS-07, Project Aura, or Spakfun ESP32 Arduino IoT Weather Station, but this project is quite advanced and the first I’ve seen based on ESP32-P4.

While you may have fun and gain knowledge and skills while building your own ESP32-P4 weather station & air quality monitor, you will almost certainly not save money compared to cost-optimized commercial products. The costs are estimated as follows for the main unit:

• Waveshare ESP32-P4-Module-DEV-KIT-C –  $77.99 on Waveshare or $98.55 on Amazon
• Heltec V4 board –  $29.99 on Amazon
• Sensirion SEN66 – $62.03 on the Sensirion website
• Light sensor BH1750 – $1.39 on AliExpress
• C4001 24GHz (12 meters) – $12.90 on DFRobot

That’s at least $185. You’ll need to add taxes and shipping, as well as a few nuts and bolts, the filament for 3D printing, and extra electronics components if you want to add external modules.

The post DIY ESP32-P4 weather station & air quality monitor connects to LoRa and ESP-NOW sensors, various weather APIs appeared first on CNX Software - Embedded Systems News.

The LightMake L4 is a desktop 3D printer with four independent printing heads for high-throughput multi-color, multi-material, and batch production. It uses four independently controlled printheads and linear-motor motion instead of conventional belts, allowing users to print four identical models simultaneously, produce complex multi-color parts with minimal purge waste, or combine up to four materials in a single print.

Typically, multi-color FDM 3D printers like the Bambu Lab X1 or Geeetech A20M use a single-nozzle purge material during every color change, which leads to a lot of material waste and longer print times. The LightMake L4 takes a different approach by keeping four filaments loaded and heated in four completely separate extruders. According to the company, this allows for one-second toolhead changes and near-zero purge waste. The linear motor system also promises ±1µm motion precision and speeds of up to 500 mm/s.

LightMake L4 and L1 specifications:

• Printing technology – FDM (Fused Deposition Modeling) / FFF (Fused Filament Fabrication)
• Storage – 4GB eMMC flash, expandable via external USB
• Display – 6.5-inch color touchscreen
• Camera
  • LightMake L4 – Dual 1560×720 @ 30 fps
  • LightMake L1 – Single 1560×720 @ 30 fps
• Connectivity – Wi-Fi,  Wired LAN  (for remote control and cluster management)
• USB – USB Type-A port for an external USB drive
• Build Volume (W × D × H)
  • LightMake L4 (Single-Color) –354 × 370 × 386 mm
  • LightMake L4 (Multi-Color)  – 354 × 350 × 386 mm
  • LightMake L1 – 395 × 395 × 386 mm (19% larger than L4)
• Motion and drive system
  • Type – Linear Motor Direct Drive with closed-loop control (±1μm precision)
  • Print Speed –  Up to 500 mm/s
  • Travel Speed – Up to 1000 mm/s
  • Acceleration – Up to 20 m/s²
  • The linear motors are rated for a 50,000+ hour lifespan
• Toolheads and extrusion
  • LightMakeL4 – 4x independent toolheads; 8x sets of linear motors
  • LightMakeL1 – 1x toolhead; 3x sets of linear motors
  • Hotend Temperature – Up to 320°C
  • Nozzle – 0.4 mm standard (0.2, 0.6, and 0.8 mm optional)
  • Cooling – Auxiliary model cooling supported
• Heated bed – Textured PEI steel sheet with max bed temperature of 110°C
• Print modes – duplicate, mirror, 4-color/4-material, plus same-material handover if one spool runs out
• Supported materials – PLA, ABS, PETG, TPU, ASA, PVA, PET, PLA-CF, PETG-CF (PA and PC supported with the optional thermal top cover)
• Misc
  • RFID material identification with automatic sync
  • Vibration cancellation in mirror/duplicate mode
  • 30 µm optical alignment for multi-material seams
  • Noise level – L4: 45 dB; L1: 43 dB (optional Thermo Top can drop noise below 40 dB)
• Power
  • LightMake L4 – 220–240V (Max 1800W); 110–120V (Max 1300W)
  • LightMake L1 – 220–240V (Max 1100W); 110–120V (Max 700W)
• Frame – Open-frame design with a monolithic die-cast frame (heated chamber available as an add-on)
• Operating temperature and humidity – 10–35°C; 20–80% RH

In terms of software and support, what I understand is that the LightMake L4 and L1 will run a customized firmware based on Klipper, with LightMake’s engineering posts showing the motion system driven through a Klipper/Fluidd stack. Model preparation is handled by LightMake Slicer (which should ship with the printers once the campaign ends), a customized OrcaSlicer build with extra support for the L4’s four independent toolheads and multi-material workflows. The company also mentioned that the slicer includes conversational AI that can apply print settings from a natural-language prompt.

The printers also include onboard cameras for unattended monitoring; the L4 has dual 1560×720 30 fps cameras, and the L1 has one camera with the same specs. LightMake says an AI monitoring system can stop common print failures mid-job. Official filament spools with RFID tags can sync material presets automatically. For multi-machine setups, one-tap fleet pairing works across about 8 meters, and the company also advertises AI job scheduling, a cluster dashboard, optional WhatsApp control through an OpenClaw integration, and OTA firmware updates.

Just last month, we wrote about the Sovol M1D hybrid IDEX 3D printer with six interchangeable toolheads, which gives the printer up to seven independently heated toolheads for different materials, colors, and support filaments. But the LightMake L4 takes a different approach, using four permanently installed independent printheads instead of interchangeable heads. This allows it to switch between heads in about one second and print four identical models at the same time for batch production. The L4 also offers a larger 354 × 370 × 386 mm build volume and uses linear motors with closed-loop positioning, with the company claiming ±1 µm motion precision. While the M1D offers greater toolhead flexibility, the L4 is more focused on faster tool changes and higher printing throughput.

The LightMake L4 and L1 printers are available on Kickstarter. The Super Early Bird pledge for the LightMake L4 is set at $1,299, a large discount from its $2,399 MSRP. The stripped-down, single-toolhead LightMake L1 is also available for $1,099. LightMake also offers various bundle options, such as the “L4 Thermo Pack” for $1,469 (which adds the Thermo Top Kit Cover), along with accessories like the Automatic Print Removal System (R1). Deliveries are expected to begin in November 2026.

The post LightMake L4 3D printer features four independent heads for simultaneous or multi-color printing (Crowdfunding) appeared first on CNX Software - Embedded Systems News.

Last December, we wrote about XyphroLabs’s UsbGpib open-source hardware, affordable, and portable USB-to-GPIB adapter designed for legacy GPIB/IEEE-488 instruments and offering modern hardware and software.

At the time, the UsbGpib V3 was also in the works with Ethernet (PoE) support, and the good news is that it’s now available, but named GPIBee. The less good news is that while V1 and V2 were both open-source hardware products (and still are), the GPIBee has been released as closed-source hardware for reasons we’ll explain below.

GPIBee specifications:

• Microcontroller – Not disclosed. But not the Microchip ATMega32U4 8-bit AVR MCU found in V2 since its USB interface is limited to 12 Mbps
• Host interfaces
  • USB 2.0 (480 Mbps) Type-C port with support for four modes
    • Full USBTMC (USB Test and Measurement Class)
    • VXI-11 over USB-Ethernet (CDC NCM)
    • linux-gpib native interface
    • Prologix-compatible (++ commands) via CDC UART (virtual COM port)
  • 10/100Mbps Ethernet RJ45 port with PoE and support for
    • VXI-11 over TCP/IP
    • Prologix-compatible (++ commands)
• 24-pin GPIB interface – Fully IEEE-488.1 and IEEE-488.2 enabled
• Misc
  • Button for factory reset (5x presses), firmware-update mode (5s hold), or fast web-interface access in USBTMC mode (single press)
  • Status LEDs (Ethernet: green, USB: orange) to indicate data activity
  • Aluminum housing
• Power Supply
  • 5V, ≤ 500 mA via USB-C or
  • 802.3af Power over Ethernet (PoE)
• Dimensions – Compact form factor
• Temperature Range, Humidity – TBD

Besides an Ethernet/PoE RJ45 port, the GPIBee also gains a more professional aluminum enclosure and faster USB 2.0 (480 Mbps) connectivity, which enables VXI-11 protocol over a USB-Ethernet gadget, concurrent sessions, and faster GPIB implementation.

Beyond the hardware, the new Ethernet/USB-to-GPIB adapter also benefits from a feature called “Benches” allowing users to create web frontends without writing code for GPIB equipment and make it available for interactive use from a web browser on desktop or phone. It’s only a front panel for instruments and can’t be used as a test automation tool, for which you can use the VXI-11 or USB interface with another program.

The dedicated website for the GPIBee mentions it’s “using open, non-proprietary standard protocols”, and “No proprietary drivers. No proprietary software.” As I understand it, the latter means “no proprietary host stack required”, and I was unable to find the latest web UI/Benches and GPIBee firmware source code.

So why isn’t GPIBee (UsbGpib V3) open-source hardware like V1 and V2? This is explained on the GitHub repository, and the key section is:

… I’ve always been and are genuinely happy to see people or companies build their own UsbGpib – that’s exactly what open source is for. What sat less well with me was seeing unrelated manufacturers turn V2 into a business of their own, selling it at scale with no attribution back to the project. That’s a different thing entirely from working with official partners like Elecrow and Binho, who I have a real relationship with and who’ve supported this project properly.

It’s never ideal when an open-source project morphs into a closed-source project for the next revision, especially if the community has contributed, but getting one’s work used without attribution is always disheartening. Having said that, the V1/V2 adapters were released under a “MIT License” that allowed commercial use.

XyphroLabs sells the GPIBee for $350 on Binho (US), and it’s also coming to a German distributor (unnamed for now) for European customers.

The post GPIBee (UsbGpib V3) USB-to-GPIB adapter gains Ethernet (PoE), but launches as closed-source hardware appeared first on CNX Software - Embedded Systems News.

Circuit Valley’s CHC5 is a modular camera system based on a Core board powered by an AMD/Xilinx Zynq-7020 SoC FPGA, a range of interface boards with USB, HDMI, or Ethernet (PoE), various sensor boards, and a metal enclosure.

The solution can be customized with different lens objectives and different sensors with MIPI D-PHY, LVDS, Sub-LVDS, Low Voltage LVDS, or SLVS interfaces. The Zynq-7020’s Cortex-A9 cores can handle complex tasks, while the FPGA fabric takes care of high-speed calculation and correction on the image, and can also be leveraged for  AI/machine-vision-specific algorithms or complex low-latency hardware acceleration.

CHC5 specifications:

• Core board
  • SoC – Xilinx AMD Zynq-7020 SoC FPGA
    • CPU – Dual-core Arm Cortex-A9 processor
    • FPGA – 85K logic cells, 4.9Mb Block RAM, 220 DSP slices
  • System Memory – 1GB DDR3
  • Storage
    • 1 Gbit flash memory
    • MicroSD card slot
  • ISP – Full-fledged ISP implemented on FPGA fabric
• Multiple interface boards with 10Gbps USB 3.0, 5Gbps USB 3.0, optional PoE, optional HDMI, etc; connect over PS and PL connectors on Core board
• Interchangeable lens mount
  • Minimum flange distance of 8.75mm
  • Currently available lens mounts: C-Mount, CS-Mount, Canon RF-Mount (Manual Controls), Sony E-Mount (Manual Controls), M42 Mount with Multiple Flange Distance, S Mount, M43 or Custom Lens Mount
• Frame Rate
  • 600 Mpixel/second ISP processing rate
  • ISP max
    • 640×480 ~2000 FPS
    • 1080p ~170 FPS
    • 4K ~40 FPS
• Supported sensors specs
  • Interface – MIPI D-PHY, LVDS, Sub-LVDS, Low Voltage LVDS, SLVS
  • Resolution – Up to about ~128MP
  • Shutter – Global or Rolling Shutter
  • Data Lanes – Up to 8 lanes
  • Number of sensors – 2x (4 lanes) or  3x (2 lanes)
  • Max data rate – 1.25Gbps/Lane, max 10Gbps with 8 lanes
  • Pixel format – RAW8, RAW12, RAW14 , PWL HDR
• Host Interface
  • USB 3.0 (5 Gbps) – USB UVC, USB3 Vision
  • Gigabit Ethernet – GigE Vision, Optional PoE
  • HDMI up to 1080p60
  • Simultaneous streaming over all 3 interfaces
• Aux I/Os
  • Isolated input, user-configurable
  • Isolated output, user-configurable
• Camera Enclosure – Works with sensor mount or no sensor mount at all, includes internal filter and M43 thread for standard external filter

The CHC5 modular camera is partially open-source hardware with PDF schematics, PCB layout, STEP files, a Linux SDK based on buildroot, and the Cypress FX3 firmware source code are all released on GitHub under a Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License. The HDL code for the ISP implementation doesn’t appear to be open source. A blog post on Circuit Valley provides more technical details about the project.

We’ve written about modular cameras in the past with products like the MaixCAM2 modular 4K AI camera or Ohm Lab Neuro N6, but that usually means just changing the camera sensor module, and the CHC5 goes much further.

If you are interested in the project, you’ll find it on Kickstarter with a 10,000 Euros funding target. You’ll need to select the $684 US CHC5 Camera System Base with the Core PCB, Interface PCB, CNC enclosure with internal filter, and your choice of camera sensor and mount. If you feel indecisive, you could always go all in with the 2,999 Euros CHC5 Developer Edition Kit Pro pictured above. Shipping adds 20 Euros to Europe, and 39 Euros to most of the rest of the world. Rewards are expected to ship in January 2027.

The post Circuit Valley CHC5 – A modular USB, HDMI, and Ethernet camera system (Crowdfunding) appeared first on CNX Software - Embedded Systems News.

Qualcomm has announced the Qualcomm Dragonwing Q-2390 and IQ-2390 quad-core Cortex-A78/A55 processors for AIoT consumer, commercial and industrial applications.

The Dragonwing Q-2390 processor is designed for commercial and consumer devices, notably for applications such as retail and smart home products, as well as smart displays and enterprise edge devices, while the Dragonwing IQ-2390 processor is the SKU from the Dragonwing IQ2 Series designed for industrial automation, machine vision, building management and energy systems.

Qualcomm IQ-2390 and Q-2390 specifications:

• CPU
  • 1x Arm Cortex-A78 core up to 1.5 or 1.9 GHz depending on SKU
  • 3x Arm Cortex-A55 cores up to 1.5 or 1.9 GHz depending on SKU
  • Optional SiFive E61 RISC-V real-time core up to 600MHz with 768kB memory
• Cache – 512 KB L3 cache
• GPU – Qualcomm Adreno 704 up to 1.1 GHz
• Audio DSP – Qualcomm Hexagon V66  up to 1.1 GHz
• NPU – Qualcomm Hexagon up to 1.1 TOPS
• DPU (Display Processing Unit) – Qualcomm Adreno 921
• VPU
  • Decode – Up to 1080p @ 30 fps H.264, H.265, VP9
  • Encode – Up to 1080p @ 30 fps H.264, H.265
• Memory – Dual-channel 16-bit LPDDR4x up to 1.8 GHz
• Storage – eMMC 5.1, SD 3.0
• Display Interface – 4-lane MIPI DSI up to 1080p60
• Camera interfaces
  • Up to 2x 4-lane MIPI CSI
  • 13 MP dual camera (ZSL, 30 fps)
  • 25 MP Single Camera (ZSL, 30 fps)
• Audio – Optional WSA8865 codec
• Networking
  • 2x Gigabit Ethernet with TSN
  • Wi-Fi 5 and Bluetooth 5.0/5.1 via WCN3950/WCN398x
  • Optional 4G LTE Cat 4
• GNSS – Optional Gen 9 VT v4-Lite GPS L1
• USB – USB 3.1 Type-C/Micro USB, 1x USB 2.0 Type-C
• PCIe – PCIe 2.0 x1
• Other I/Os – 10x QUP, 155x GPIO, UART, I²C, SPI
• Dimensions
  • Dragonwing Q-2390 – 12 x 12 x 0.92 mm; 0.4mm ball pitch
  • Dragonwing IQ-2390 – 21.3 x 18.4 x 1.82 mm; 0.8mm ball pitch
• Temperature Range
  • Q-2390 AA variants -30°C to +95°C
  • Q-2390 AB variants and IQ-2390 – -30°C to +115°C
• Longevity  – Available until 2036
• Part Numbers
  • Dragonwing Q-2390 – CQ2390S-0-AA, CQ2390S-0-AB, CQ2390M-0-AA, CQ2390M-0-AB; AA vs AB variants: AB gets NPU, RISC-V core, and Hexagon V66 audio DSP; S vs M parts: M gets 4G LTE and GNSS
  • Dragonwing IQ-2390 – IQ2390-AB (no 4G LTE modem, no GNSS)

Qualcomm provides support for Android, Yocto Linux, and Ubuntu on the Cortex-A application cores, and  Zephyr RTOS for the parts with a RISC-V real-time core.

The company will be showcasing Dragonwing Q-2390 and IQ-2390 SoMs (System-on-Modules) at IFA 2026, and evaluation kits are expected to become available in Q1 2027. Additional information may be found on the product pages for the new processors and in the press release.

The post Qualcomm introduces Dragonwing Q-2390 and IQ-2390 for consumer and industrial AIoT applications appeared first on CNX Software - Embedded Systems News.

The Makera Z1 Desktop CNC is a desktop CNC machine designed for milling and engraving work that requires precision. The fully enclosed machine features an aluminum structure, offers a 200 × 200 × 100 mm work volume, and is suitable for wood, plastic, acrylic, PCB boards, carbon fiber, as well as non-ferrous metals such as aluminum, brass, and copper.

The standout feature of this model is its fairly complete set of assistive functions, including a quick tool change system for swapping milling bits, auto probing & leveling for detecting position and leveling the workpiece, the AeroDust system for blowing and managing chips/dust, and a built-in camera for monitoring operation. In addition, optional accessories such as a 4th-axis rotary module, a 5W laser head, a workholding kit, and a 3D probe can also be installed.

This model relies on a linear rail system featuring Acme lead screws and NEMA 17 stepper motors, supporting a motion resolution of about 0.02 mm. It comes with a 150W spindle with a maximum rotation speed of 13,000 RPM, suitable for small CNC jobs that require detail and consistent operation.

Makera Z1 Desktop CNC Specifications

• 3-axis work area – 200 × 200 × 100 mm
• 4-axis work area – 80 mm diameter × 150 mm length (optional)
• Clearance under the gantry – 115 mm
• Spindle power – 150W
• Spindle speed – 0–13,000 RPM with closed-loop speed control
• Standard collet size – 1/8 inch
• Additional supported collet sizes – 1/4 inch, 6 mm, 4 mm, and 3 mm
• Tool change system – Quick tool change
• Spindle cooling – Built-in air cooling
• Drive motors – NEMA 17 Stepper Motors
• Motion system – 12 mm Linear Rails with Acme Lead Screw
• Maximum travel speed – 1,200 mm/min
• Motion resolution – 0.02 mm
• Spindle runout – 0.01 mm
• Repeatability – 0.002 mm
• Probing and leveling – Auto probing & leveling
• Chip and dust management – AeroDust collection system
• Camera – built-in camera for monitoring operation and recording time-lapse footage
• Laser module support – 445 nm 5W diode laser (optional)
• 4th-axis support – Belt-driven system (optional)
• Maximum workpiece weight on the table – 5 kg
• Operating noise level – about 70 dB, depending on material and bit
• Supported materials – MDF, plywood, plastics, acrylic, PCB/FR4, carbon fiber, aluminum, brass, copper, and other non-ferrous metals
• Software – Makera Studio and Makera App
• Supported CAD/CAM software – Fusion 360, SolidWorks, AutoCAD, VCarve Pro, and Aspire
• Machine size – 350 × 470 × 450 mm
• Height with the lid open – 840 mm
• Weight – About 17 kg

What’s in the Box

The box contains the following items:

• Makera Z1 Desktop CNC machine
• Wired Probe kit for detecting position and setting Z-axis height
• Emergency stop button
• Starter milling bit set
• Collet set for holding bits of various sizes
• Tools for installing and removing collets
• Clamp and screw set for workholding
• Locating pins to help position the workpiece
• Multi-purpose screwdriver and hex bits
• Power cable and power adapter
• USB Type-C cable for connecting to a computer
• Safety glasses
• Lubricating oil for machine maintenance
• Double-sided tape for holding workpieces
• Sandpaper block for finishing workpieces
• Sample materials for test cuts
• Quick start guide, safety guide, and parts list

Each item is packed separately, and there is a QR code for immediately opening installation or usage instructions on the Makera Wiki. However, accessories such as the 4th-axis module, 5W laser head, 3D wired probe, low-profile vise, vacuum bed, and cyclone dust collector are optional.

The Makera Z1 supports a fairly wide range of materials. The kit we received for testing included several types of sample sheets, such as clear acrylic, colored acrylic, plastic, synthetic material for 3D forming, as well as aluminum and brass sheets, for testing 2D milling, letter engraving, 3D relief, and small metal milling.

Each material has different properties, so the bit, spindle speed, feed rate, and depth per pass must be selected accordingly. The same cutting values should not be used for every material. Even though Makera Studio provides default values and tool recommendations, it is still best to check and test on scrap material before starting the real job.

Materials Supported by the Makera Z1

• Wood and wood products such as softwood, hardwood, MDF, and plywood, suitable for engraving, signs, decorative items, and low-relief work.
• Plastics such as ABS, PVC, acrylic, PC, and HDPE, suitable for signs, prototype parts, covers, and robot parts.
• Synthetic modeling materials, suitable for testing toolpaths, 3D relief work, and 4-axis sculptures because they cut easily and retain detail fairly well.
• PCB and FR4, supporting circuit milling, drilling, and board outline cutting. Auto Leveling should be used to compensate for uneven surfaces.
• Carbon fiber can be milled, but dust control must be strict because carbon fiber dust is hazardous to the respiratory system and can be electrically conductive.
• Aluminum is suitable for engraving, signs, and small parts. Carbide bits should be used, and cutting should be done in multiple shallow passes.
• Brass and copper can be milled, but feed rate, step-down, and chip evacuation must be adjusted properly to prevent surface scratches or chips packing around the bit.

From our tests, soft materials such as wood, plastics, and synthetic materials were easy to mill and produced consistent results. Aluminum and brass can also be milled, but they take longer and require more careful settings, especially when using small bits or highly detailed toolpaths. The recommendation is not to cut aluminum deeper than about 1 mm per pass, and with a 150W spindle and small bits, we recommend starting at about 0.1–0.3 mm per pass and then increasing according to the material, bit, and toolpath. This helps reduce cutting force and the risk of missed steps.

The Makera Z1 is not designed for milling steel, stainless steel, or titanium as a primary use case, and it is not suitable for materials that require wet cooling, such as glass or stone.

For the 5W laser head, material requirements should be treated separately from CNC milling. The laser is suitable for wood, leather, fabric, paper, and some plastics, but it should not be used on PVC or vinyl since those materials can release toxic gases and corrode the machine. Clear acrylic should be cut with a CNC bit because a 445 nm diode laser cannot cut it.

Makera Cyclone Dust Collector Lite

This is a dust extractor and separator designed specifically for use with desktop CNC machines. It uses a motor with a maximum power of 200W and a cyclone dust-separation system combined with multi-stage filtration. It extracts wood chips, plastic, acrylic, and metal debris generated during milling from the work area to reduce debris buildup inside the Makera Z1.

The cyclone system spins the airflow to separate larger debris from fine dust. It can separate about 97% of the larger debris into the collection bin before the air enters the filter assembly, helping reduce clogging and extending filter life.

Inside the unit is a 3-stage filtration system consisting of a pre-filter sponge, a HEPA filter, and an activated carbon filter for capturing fine dust and helping reduce odors from milling or engraving. There is also a 6-liter collection bin at the bottom, and the debris level can be checked from the outside.

It is recommended to use about 60–70% suction for wood and plastics, and increase it to 80–100% for metal milling. The filter assembly should also be checked and cleaned after about 30 hours of accumulated use. The actual interval depends on the material type, debris volume, and operating time.

When the top cover of the Makera Cyclone Dust Collector Lite is opened, the air filter compartment and dust filter assembly can be seen inside. The system extracts debris from the Makera Z1 through a hose, then drops most of the debris into the lower collection bin while fine dust is trapped by the filters before the air is exhausted. The openable top cover makes it convenient to inspect and clean the filters, which is important for maintaining suction and performance. In our testing, the filter section was easy to access without having to disassemble the entire unit.

Makera Cyclone Dust Collector Lite features

• Maximum motor power of 200W
• Cyclone separation system
• 3-stage filtration: pre-filter, HEPA, and activated carbon
• Manufacturer-rated filtration efficiency up to 99.99%
• Maximum vacuum pressure of 12 kPa
• Maximum airflow of 900 L/min
• 6-liter debris collection bin
• Maximum noise level of about 70 dB
• Suction level adjustable via a front rotary knob
• Supports on/off and suction adjustment from the Makera App when connected to the CNC machine
• Supports automatic operation via G-code commands, starting and stopping together with the CNC machine
• Has a manual control mode for standalone use
• Connects to the Makera Z1 through a suction hose and control cable
• Can be used with the Vacuum Bed to create suction for workholding
• Body made from aluminum alloy and plastic parts
• Dimensions – 300 × 218 × 354 mm
• Included accessories consist of a suction hose, filters, connection cable, power cable, and manual

The rear of the Makera Cyclone Dust Collector Lite has the main connection points. The upper left is the debris suction hose port for connecting the hose between the extractor and the Makera Z1 so wood, plastic, acrylic, or metal chips can enter the cyclone separation system.

The lower right is the AC power inlet with an ON/OFF switch for powering and turning the unit on and off. The port labeled Control is used to connect a control cable to the Makera Z1 so the CNC machine can automatically start, stop, or adjust the extractor in relation to the milling process. It also supports control through the Makera App when the extractor is connected to the CNC machine.

The rear also has a ventilation opening to help cool the motor and internal electronics. During installation, enough space should be left behind the unit. It should not be placed tightly against a wall or have the vent blocked. The suction hose, power cable, and control cable should also be checked to make sure they are fully seated before use.

This shows the rear connections of the Makera Z1 Desktop CNC when used with the Makera Cyclone Dust Collector Lite. The main suction hose carries debris from inside the CNC machine to the extractor, while the Control cable sends signals between the devices so the extractor can start or stop automatically according to the Makera Z1 milling process.

The rear panel of the Makera Z1 CNC machine includes the following:

• Power inlet and on/off switch
• Debris suction hose port for connecting the machine’s internal AeroDust system to the Cyclone Dust Collector Lite
• Control/Data port
• Emergency stop port
• USB Type-C for connecting the Makera Z1 to a computer and for network setup
• MicroSD card slot for storing and transferring job files

Once the control cable and suction hose are fully connected, the extractor can work in sync with the Makera Z1. This reduces the need to manually turn the extractor on and off. However, the hose should be routed so it is not kinked, folded, or bent too tightly, as that can reduce airflow. Space should also always be left around the ventilation openings of both devices.

Makera Z1 Emergency Stop Button

The Emergency Stop is an external stop button connected to the Makera Z1. It is used to immediately stop the machine when an abnormality is found, such as the bit hitting a fixture, the workpiece coming loose, an incorrect toolpath setting, or unusual noise and vibration.

Makera Z1 150W Spindle Head

The Makera Z1 uses a 150W spindle head supporting speeds from 0–13,000 RPM, with closed-loop speed control to keep RPM consistent during milling. It is suitable for wood, plastics, acrylic, PCB, carbon fiber, and non-ferrous metals such as aluminum, brass, and copper. The bit, RPM, feed rate, and depth per pass must still be chosen according to the material.

Spindle runout is about 0.01 mm, which helps reduce bit wobble and supports jobs that require great detail. The spindle uses internal air cooling, with a cooling vent at the top of the head and a blue LED strip on the front to indicate operating status.

Bit holding is handled by a purpose-designed collet. It supports 1/8-inch bits as standard and can also be used with 1/4-inch, 6 mm, 4 mm, and 3 mm collets. The part I like most is the Quick tool change system, which makes it faster to remove or swap milling bits and the wired probe without having to tighten the collet with a wrench through multiple steps like on a typical CNC machine. It works by flipping the lever up and down.

Makera Wired Probe

This is a wired probe that installs into the spindle head in place of a milling bit. It is used to detect the surface, edges, and reference position of the workpiece before milling begins. This reduces the need to set axes manually and reduces error from visual measurement.

Installation starts by stopping the spindle, inserting the wired probe into the quick tool change system, and plugging the signal cable into the port on the side of the machine head. Once installed, the software commands the probe to move until it touches the workpiece surface or edge in order to calculate the reference position and automatically adjust the work offset.

The wired probe can also be used with auto probing & leveling to survey surface height at multiple points, especially for PCB work or thin sheet materials whose surfaces may not be perfectly flat. The system then uses the measured values to compensate for level during milling, helping keep groove depth more consistent.

Makera 3D Wired Probe

This optional accessory is used to detect workpiece position in the X, Y, and Z axes. The kit includes a 3D probe head with a ruby-tipped stylus, a connection cable, 1/8-inch, 4 mm, 6 mm, and 1/4-inch stylus adapters, plus a hex wrench and installation tools. The probe can find edges, measure surfaces, determine center points, and automatically set work offsets. This helps reduce measurement error from manual methods, especially on jobs that require flipping or repositioning the workpiece. The 3D wired probe is a measuring device, and the spindle must not be turned on while it is installed.

Installing a Milling Bit with the Quick Tool Change System

To rapidly remove and install milling bits, the user lifts the unlock lever on the side of the spindle head, inserts the bit assembly into position, and then presses the lever back down to lock the tool securely. This system reduces the need to tighten and loosen a collet with a wrench every time a bit is changed.

Quick tool change is useful for jobs that require several bit types in sequence, such as starting with a roughing bit to remove material and then switching to a smaller bit for detail work. It also makes it easier to switch from a milling bit to the Wired Probe for measuring workpiece position.

While it is not an automatic tool changer that picks and swaps bits by itself, it still significantly reduces time and hassle during the workflow. In our testing, unlocking and inserting a bit was easy and did not require extra tools. Before installing or removing a bit, wait until the spindle has fully stopped, and check that the bit assembly is inserted all the way and that the lever has returned to the locked position. If it is not seated securely, the bit may vibrate, come loose, or damage the workpiece during operation.

Makera Low-Profile Vise

This workholding vise uses side clamping force to hold material in place, unlike top-down clamps that can obstruct the bit path. The vise can be adjusted to suit the workpiece size and has numbered tightening points to help users install and lock the workpiece in sequence. Its low-profile design opens up space above the workpiece and reduces the risk of the spindle head or bit colliding with the fixture during movement.

In our testing, the low-profile vise held square stock and small workpieces firmly while leaving enough space for the bit to access the part. However, clamping force should be checked so it is not too tight on brittle materials, and the machine head should be jogged over the work area before milling starts to confirm that the toolpath will not collide with the vise.

Makera Studio software and Makera App

Makera Studio for Windows and macOS combines CAM and machine control in a single program. Users can import models, select bits, generate and simulate toolpaths, control axes, set the start point, and send jobs directly to the machine. The Makera App for Android and iOS is used to check status, progress, and the built-in camera feed from a smartphone. However, the version of Makera Studio we tested was still in public beta, so some functions and interface elements may change in the future.

Makera Studio includes Machining Mode options suited to different job types, including 2D Vector Machining for cutting or engraving paths, 3D Model Machining for forming 3D models, 3-Axis Relief Carving for low-relief images, and a laser mode. There are also Custom 3-Axis Machining and Custom 4-Axis Machining modes for users who want to define bits, toolpaths, and operating values themselves. Guided Mode is suitable for beginners, while Custom Mode offers more setup freedom.

The Device Control page in Makera Studio is used to control and monitor the Makera Z1. It displays a toolpath preview, the positions of the X, Y, Z, and A axes, and Jog buttons for moving axes, setting the origin, and calling the 3D Probe or Auto Probe. Users can also open files, start or stop jobs, and check progress, elapsed time, and remaining time from a single screen. This makes it possible to go from inspecting the toolpath to sending the job to the machine within one program.

Milling and Engraving Tests with the Makera Z1

We started testing the capabilities of the Makera Z1 on various materials including soft plastic, aluminum, and brass.

Letter Milling Test on Soft Plastic Sheet

We tested the Makera Z1 by milling and engraving text onto a soft plastic sheet using a 2-flute straight end mill, 1/8 inch or 3.175 mm in diameter. This type of bit is suitable for slot milling and lettering that needs a relatively flat groove floor.

The approximate cutting values used in the test were as follows:

• Spindle speed: about 12,000 RPM
• Feed rate: about 500 mm/min
• Plunge rate: about 150 mm/min
• Step-down: 0.5 mm
• Total letter depth: about 1 mm
• Number of passes: 2
• Runtime: about 7–10 minutes, depending on the toolpath and letter detail

Before starting the job, we used the wired probe to adjust the position and set the workpiece surface level. We then created a toolpath in Makera Studio using pocket milling to remove material inside the letters. The machine cut down about 0.5 mm at a time for a total of 2 passes. This helped reduce cutting force on the bit and produced cleaner letter edges than cutting the full depth in a single pass.

The result showed that the machine could mill straight lines, curves, and the interior areas of the letters fairly clearly. The groove floor was consistent, and the workpiece did not shift during the test. However, the inside corners of the letters had a radius matching the bit, which is a natural limitation of milling with a cylindrical cutter. If smaller letters or sharper corners are needed, a 1–2 mm bit or a V-bit designed specifically for letter engraving can be used instead.

For the next test, we milled a smaller “CNX Software” text on a soft synthetic sheet to evaluate the Makera Z1’s accuracy in creating straight lines, curves, interior letter spaces, and closely spaced details. The workpiece was held with the Low-Profile Vise to prevent movement during milling.

For this type of job, a 2-flute flat end mill about 1 mm in diameter can be used. It is small enough to reach narrow spaces inside letters while still creating a flat groove floor, unlike a V-bit, which creates a V-shaped groove whose width changes with depth.

The cutting values for the test were as follows:

• Spindle speed: about 12,000–13,000 RPM
• Feed rate: about 250 mm/min
• Plunge rate: about 80 mm/min
• Step-down: about 0.25 mm
• Total depth: about 0.5 mm
• Number of passes: 2
• Toolpath: 2D pocket plus a contour pass for edge cleanup
• Runtime: about 10–15 minutes, depending on text size and the number of toolpaths

Against, we used the wired probe before milling. The machine then cut down about 0.25 mm at a time to reduce load on the small bit and lower the risk of breakage. After milling the interior of the letters, one additional contour pass can be added to clean the edges and make the letter shapes clearer.

The result showed that the small letters were still readable. Vertical lines and curves remained continuous, and most interior letter spaces stayed clearly separated. However, inside corners had a 0.5 mm radius matching the bit, which is a physical limitation of a cylindrical cutter. If smaller letters or sharper corners are needed, a 0.5–0.8 mm bit or a 30-degree V-bit can be used, but feed rate and step-down must be reduced to prevent bit breakage.

High-Detail Relief Milling Test with the Makera Z1

We tested the Makera Z1’s fine milling capability using a 3D relief wolf model. The model was imported into Makera Studio to generate the toolpath. The software made toolpath setup fairly convenient, but because the model contained a large amount of detail, including fur, the face, eyes, ears, and trees in the background, milling took noticeably longer than lettering or 2D cutting.

Jobs like this are split into two stages: roughing to remove most of the material, and finishing to refine the surface and model details. Approximate settings for the workpiece in the photo were as follows:

Roughing

• Bit: 3.175 mm two-flute flat end mill
• Spindle speed: about 12,000 RPM
• Feed rate: about 500 mm/min
• Plunge rate: about 150 mm/min
• Depth per pass: about 1 mm
• Stepover: about 40% of bit diameter
• Approximate time: 45–90 minutes

Finishing

• Bit: approximately 1 mm ball nose end mill
• Spindle speed: about 13,000 RPM
• Feed rate: about 250–300 mm/min
• Stepover: about 0.08–0.12 mm
• Toolpath: 3D parallel finishing or raster finishing
• Approximate time: 7–10 hours

When combining roughing, bit changes, probing, and detail finishing, this job may take about 8–12 hours in total, depending on workpiece size, stepover, feed rate, and the detail level selected in Makera Studio.

The main reason it takes so long is that finishing requires a small ball nose bit to pass across the surface in very closely spaced lines. If stepover is set to about 0.1 mm, the bit has to travel across the workpiece hundreds or even thousands of times. Reducing stepover further makes the surface and details smoother, but milling time increases substantially as well.

The result shows the details of the fur, facial shape, eyes, nose, and background quite clearly, especially in areas with only small height differences. However, if shorter runtime is needed, stepover can be increased to about 0.15–0.20 mm, or a 1.5–2 mm ball nose bit can be used instead. Small details and surface smoothness will decrease accordingly.

These values are approximate settings suitable for the workpiece shown in the photo. We did not record the exact start and end times or every value from the Makera Studio file because the job took so long, so they may differ slightly from the software estimate and actual runtime.

Aluminum and brass milling on the Makera Z1 Desktop CNC machine

The Makera Z1 supports milling non-ferrous metals such as aluminum, brass, and copper. However, with a 150W spindle, it is better suited for shallow milling, engraving, and small metal parts than heavy cutting or large material removal. The manufacturer recommends not cutting aluminum deeper than about 1 mm per pass. For small bits and detailed work, we used a much shallower depth to reduce cutting force and the risk of bit breakage.

In this test, we milled two small circular workpieces using the same “CNX Software” text on aluminum and brass in order to compare detail retention and surface characteristics of the two materials.

Approximate settings for this job were as follows:

• Bit type: single-flute solid carbide flat end mill
• Bit size: about 1 mm
• Spindle speed: 12,000–13,000 RPM
• Toolpath: 2D pocket plus contour finishing
• Total pattern depth: about 0.3–0.5 mm
• Cyclone Dust Collector set to about 80–100% to help remove metal chips from the cutting area

Approximate aluminum settings

• Feed rate: about 200–250 mm/min
• Plunge rate: about 50–60 mm/min
• Step-down: about 0.10–0.15 mm
• Runtime: about 30–45 minutes

Approximate brass settings

• Feed rate: about 150–200 mm/min
• Plunge rate: about 40–50 mm/min
• Step-down: about 0.08–0.10 mm
• Runtime: about 40–60 minutes

The test showed that the Makera Z1 could mill the letters and cut the circular outline on both materials. The aluminum workpiece produced letters that were relatively clear and had a more even overall surface. The brass workpiece showed more visible tool marks and a rougher surface, especially in the background and small letter areas.

This difference does not mean the machine cannot mill brass. It shows that brass requires more careful adjustment of bit choice, feed rate, step-down, and stepover, as well as continuous chip evacuation. If chips recirculate under the bit, they can be recut and leave marks on the surface.

If a smoother brass finish is needed, the job should be split into roughing and finishing passes, step-down reduced to about 0.05–0.08 mm, and a light finishing pass added. For letters that need sharper corners, a V-bit or pointed engraving bit can be used instead, but feed rate must be reduced and extra care taken to avoid breaking the tip.

Based on the test results, the Makera Z1 is suitable for engraving metal signs, prototype coins, nameplates, PCBs, and small aluminum or brass parts. It is not suitable for stainless steel or deep, fast metal removal like you would do with an industrial CNC machine.

Installing the Vacuum Bed on the Makera Z1

This is my favorite workholding base because it holds the workpiece with vacuum suction. It is suitable for flat sheet materials such as wood, plastic, acrylic, PCB boards, and thin metal sheets. It reduces the need for clamps and opens up the top of the workpiece, so the bit can move without fixtures blocking the toolpath.

Installation starts by removing clamps or the low-profile vise from the machine table. The vacuum bed is then aligned with the reference positions on the Makera Z1 base and fastened evenly with screws. The base has a grid of air channels distributed across the surface to help create suction under the workpiece when connected to a vacuum source.

After installing the vacuum bed, we placed the included wood sheet onto the grid-lined spoilboard. The numbers and grid lines helped align the workpiece with the machine’s X and Y axes. Getting the alignment right from the start makes it easier to set the reference point in software and reduces the chance of the toolpath going outside the material.

Before placing the workpiece, both the vacuum bed surface and the underside of the wood sheet should be cleaned up, so dust or chips are not trapped underneath. Debris can prevent the workpiece from sitting flat and create air leaks. If the wood sheet is warped, rough, or highly porous, the suction used to hold it may also be reduced.

After installing the vacuum bed and placing the spoilboard, the next step is connecting the vacuum hose to the air inlet on the side of the base. The clear hose included in the kit should be pushed fully onto the fitting to prevent it from coming loose and to reduce air leaks during use.

This hose carries air from the grid channels under the workpiece to the vacuum source. When the Makera Cyclone dust collector lite is turned on, the air pressure under the material drops, so outside air pressure presses the workpiece against the vacuum bed and holds it in place without clamps.

The CNC machine and extractor should be turned off before connecting the hose. Then check that the hose end is not clogged with dust, torn, or twisted. After connecting it, route the hose so it is not bent too tightly or pinched by the machine cover, as that would reduce airflow and weaken suction at the bed.

A hose and valve can be found inside the machine to set the vacuum path. When using the vacuum bed, the clear hose from the bed must be connected to the internal fitting so suction from the Makera Cyclone dust collector lite can be sent to the air channels under the workpiece.

Before switching hoses, stop the Makera Z1 and turn off the extractor. Then push the hose fully onto the fitting and set the valve to the vacuum bed position. This changes the airflow path from internal chip extraction to creating suction at the workholding bed.

After connecting it, check that the hose is not loose, kinked, or in the travel path of the table or spindle head. Then turn on the extractor and test holding force by gently trying to move the workpiece. If it still slides, recheck the valve position, connections, and possible air leaks.

One observation from testing is that when suction is redirected to the vacuum bed, chip extraction around the bit may decrease depending on the valve setting, because the same vacuum source is being used to hold the workpiece. Therefore, cutting values that generate too many chips should be avoided, the work area should be checked periodically, and debris should be cleaned up after the job.

When returning to clamp or low-profile vise workholding, turn off the extractor first, disconnect the vacuum bed hose, and set the valve back to the normal chip-extraction path. Do not pull or switch hoses while the extractor is running, as that can scatter dust and chips inside the machine.

Vacuum bed test with a plastic sheet

We then tested the system with a two-color engraved plastic sheet. The material was placed on the spoilboard and held with vacuum suction instead of clamps.

This material has a red surface layer and a white lower layer. When the bit removes the top layer, white “CNX Software” letters appear. It is suitable for testing both milling accuracy and the vacuum bed’s ability to hold thin sheet material.

Approximate milling values for the test were as follows:

• Bit type: two-flute flat end mill
• Bit size: about 3.175 mm
• Spindle speed: about 12,000 RPM
• Feed rate: about 350–450 mm/min
• Plunge rate: about 100 mm/min
• Depth per pass: about 0.2–0.3 mm
• Total depth: about 0.3–0.5 mm, or just enough to pass through the red layer
• Toolpath: 2D pocket
• Runtime: about 10–15 minutes

Before milling, we turned on the Makera Cyclone dust collector lite to create suction, then tried moving the sheet by hand to confirm that the workpiece would not shift. We then used the wired probe to measure surface position and set the reference point before sending the file from Makera Studio to the machine.

During the test, the vacuum bed held the sheet in place without clamps blocking the toolpath, allowing the letters in the center of the sheet to be milled continuously. The result had straight letter lines and no offset marks caused by workpiece movement, showing that the suction was sufficient for shallow milling on flat sheet material.

However, the vacuum bed is best suited to jobs with relatively low cutting force, such as engraving, PCB signs, and thin plastic sheets. For deeper cuts, large bits, or high side force, feed rate should be reduced, the cut should be split into multiple passes, or another workholding method should be used. Unused areas of the bed should also be covered to reduce air leaks and keep suction under the workpiece.

Makera Z1 4th Axis Module

The Makera Z1 4th axis module is an optional accessory that adds a rotary axis, or A axis, to the Makera Z1. This allows the machine to mill around cylindrical workpieces instead of only flat 3-axis surfaces. It is suitable for sculptures, cylindrical parts, handles, rings, stamps, models, and patterns that require detail around the workpiece.

The 4th-axis kit consists of a rotary head with a 3-jaw chuck, a linear rail base, and a tailstock to support the workpiece. During operation, Makera Studio coordinates A-axis rotation with cutter movement so the bit can continuously reach the surface around the workpiece without stopping the machine to flip the material by hand.

Technical Information

• Maximum work area: 80 mm diameter × 150 mm length
• Drive system: belt drive
• Rotary axis torque: about 2.5 N·m
• Workholding: self-centering 3-jaw chuck
• Support: tailstock with adjustable center
• Controlled rotary axis: A axis
• Connection: power and signal cables connected to an internal Makera Z1 port
• Software: 4-axis project creation supported in Makera Studio
• Job types: rotary roughing, rotary finishing, and wrap-around relief milling
• Installation: the module is mounted to the machine table using designated holes and reference positions
• Compatibility: designed for the Makera Z1 and not interchangeable with the 4th Axis Module from the Carvera or Carvera Air

Installing a workpiece on the 4th axis module starts by inserting the end of the material into the chuck. The included wrench is then used to turn the chuck mechanism, so all three jaws move toward the center and clamp the workpiece evenly. In the photo, we used two tools to help hold and tighten the mechanism so the chuck would not rotate while locking.

Insert the workpiece deep enough into the chuck and leave only as much overhang as necessary, because a part that sticks out too far is more likely to vibrate or deflect. Then slide the tailstock to the other end and adjust the center so it contacts the midpoint of the workpiece to help support the axis during rotation.

Chuck clamping force must be high enough to prevent the workpiece from slipping or spinning inside the jaws, but it should not be too tight, especially on soft materials such as wood, foam, or synthetics, because the jaws can deform or crack the material. The tailstock should press just enough to stop the workpiece from wobbling, but not so tightly that it adds extra friction and overloads the 4th-axis motor.

After clamping the workpiece, slowly rotate the chuck by hand through one full turn to check that the part is centered, not wobbling, and not colliding with the base or internal machine components. Then test rotation again with a low-speed jog command before starting the real cut.

In our tests, the chuck and tailstock held small cylindrical workpieces securely. However, 4-axis accuracy depends heavily on centering. If the workpiece is even slightly tilted or off-center, cutting depth around the part may be uneven and some details on one side may be lost.

After mounting the 4th axis module on the Makera Z1 table, the rotary axis cables must be connected to the port inside the machine for power and control signals. The Makera Z1 should be turned off and allowed to come to a complete stop before plugging in the cable. Then align the connector with the port slot and insert it fully without forcing or twisting the plug. If it does not go in, remove it and check the orientation again, because pressing the connector in the wrong way can damage the internal pins.

After connecting it, route the cable along the side of the work area and leave enough slack for table movement and chuck rotation. The cable must not cross the linear rails, touch the workpiece, or come near the bit, because 4th-axis rotation can pull the cable into the chuck assembly.

When the machine is powered on and 4-axis mode is selected in Makera Studio, the system sends commands to the rotary motor according to the defined toolpath. Before actual milling, use jog commands to slowly rotate the chuck through at least one full turn to check rotation direction, the connection, and cable clearance, and to confirm that the workpiece does not collide with the spindle head, tailstock, or internal frame.

In our review, the connection point was inside the machine and not difficult to reach, but the cable should be plugged in before finishing the placement or adjustment of the 4th-axis kit. Once the chuck and workpiece are in position, there is less room to reach the port.

4-Axis milling with epoxy

The sample workpiece was a small sculpture with detail on all sides, so the 4th axis had to rotate the part while the spindle head moved in X and Z. This allowed the bit to continuously reach the front, back, and sides without stopping the machine to flip the workpiece by hand.

Jobs like this can be split into two stages as follows:

Roughing

• Bit: 3.175 mm two-flute flat end mill
• Spindle speed: about 12,000 RPM
• Feed rate: about 400–500 mm/min
• Plunge rate: about 100–150 mm/min
• Depth per pass: about 0.8–1 mm
• Stepover: about 35–40% of bit diameter
• Runtime: about 45–90 minutes

Finishing

• Bit: approximately 1 mm ball nose end mill
• Spindle speed: about 13,000 RPM
• Feed rate: about 250–300 mm/min
• Stepover: about 0.08–0.12 mm
• Toolpath: rotary finishing or 4-axis parallel finishing
• Runtime: about 5–8 hours

When combining probing, bit changes, roughing, and detail finishing, a sculpture of this type may take about 6–10 hours in total, depending on workpiece size, model complexity, and the stepover set in Makera Studio.

The test showed that the Makera Z1 could create a continuous form around the workpiece, including facial details, hair, clothing, and the sculpture base, without visible seams from flipping the part as in multi-sided 3-axis milling. However, 4-axis work requires more attention to setup than 3-axis work, especially aligning the chuck and tailstock on the same centerline. If the workpiece is even slightly off-center, cutting depth on each side may be unequal.

Before starting the job, rotate the 4th axis through one full turn and check that the workpiece does not collide with the table, spindle head, or internal components. The chuck and tailstock must also be tight enough, but not so tight that the material cracks or deforms.

The technical values above are approximate settings suitable for the workpiece in the photo. If shorter time is needed, a larger ball nose bit or increased stepover can be used, but surface detail will decrease accordingly.

Working with Makera Studio

Makera Studio has a project mode for 4-axis work. Users can import a 3D model, define stock size, select the rotary axis, and generate roughing and finishing toolpaths. The software calculates workpiece rotation in relation to bit position.

High-detail jobs should be split into two stages:

• Rough with an approximately 3.175 mm flat end mill to remove most of the material
• Finish with an approximately 1–2 mm ball nose end mill to refine curved surfaces and small details

Reducing stepover in the finishing stage makes the surface smoother and details clearer, but it increases the number of toolpaths and runtime. A small sculpture with detail all the way around may take many hours.

Testing the Makera Z1 with the 5W Laser Head

The Makera Z1 can be fitted with a 5W laser module to add engraving and thin-material cutting capability. It’s a blue laser diode with a 445 nm wavelength suitable for engraving wood, leather, paper, cardboard, fabric, and some plastics.

Makera Z1 Laser Module Specs

• Laser type: semiconductor diode laser
• Laser power: 5W
• Wavelength: 445 nm
• Use cases: line engraving, filled images, and thin-material cutting
• Installation: mounts into the spindle head through the Quick tool change system
• Connection: control cable plugs into the 3-pin port on the side of the spindle head
• Software: laser job preparation supported in Makera Studio
• Supported materials: wood, some plastics, fabric, leather, and cardboard
• Included accessories: 5W laser head, connection cable, air tube, and laser safety glasses

Installation Steps

Before installation, stop the Makera Z1 and wait until the spindle has fully stopped. Then remove the milling bit and insert the laser head into the same position using the Quick tool change system. During installation, the module must be aligned so it does not rotate or tilt, because the laser head orientation and cable position must face the designated side.

After locking the laser head, plug the module’s control cable into the 3-pin port on the side of the spindle head and connect the air tube according to the manual. Check that the cable and tube do not droop near the laser head or cross the machine’s travel path.

In our testing, installation was quick thanks to the Quick tool change system and did not require removing the spindle assembly from the machine. However, before starting a job, the head should be jogged to confirm that the cable is not taut and does not contact the workpiece.

The 5W laser head is suitable for engraving logos, letters, black-and-white images, and patterns onto wood surfaces. Mark intensity depends on laser power, travel speed, and the number of passes.

Approximate starting values for test engraving on wood are as follows:

• Laser power: about 60–80%
• Speed: about 1,500–2,500 mm/min
• Number of passes: 1
• Line interval for filled images: about 0.10–0.15 mm

For cutting thin wood or cardboard, power close to 100%, lower speed, and multiple passes may be required. Cutting capability depends on the material type, thickness, density, and moisture, so scrap material should be tested every time.

Limitations of the 5W Laser Head

This laser module is suitable for engraving and thin-material cutting. It cannot replace a high-power laser machine. The manufacturer states that it is not suitable for cutting acrylic sheets with the laser, especially clear acrylic, which does not absorb 445 nm blue light well. If acrylic needs to be cut, a CNC bit should be used instead.

Safety

• Wear the included laser safety glasses every time during installation, testing, and use.
• Close the Makera Z1 cover before starting engraving.
• Do not look directly at the beam or focal point.
• Do not leave the machine running unattended, because materials can catch fire.
• Turn on the extraction and air filtration system to remove smoke, odor, and particles.
• Do not use it on PVC, vinyl, or chlorine-containing materials, as they can release toxic gases and corrode the machine.
• Check the material type before use. If the composition is unknown, do not laser it.
• Keep suitable fire-extinguishing equipment near the work area.

Overall, the 5W Laser Module lets the Makera Z1 handle both milling and laser work in one machine. It is suitable for signs, souvenirs, artwork, and small prototypes, but it should be used within the laser’s power limits and with strict attention to safety requirements.

Wood engraving with the 5W laser module

We tested the Makera Z1’s 5W diode laser module by engraving a landscape image onto an MDF sheet about 100 × 100 mm. The source image had a fairly large amount of detail, including mountains, pine trees, a house, a lake, and rock textures, making it suitable for evaluating resolution, intensity control, and the laser head’s ability to create fine lines.

Before starting the job, we imported the image into Makera Studio, converted it to black-and-white or grayscale, and generated a raster engraving toolpath so the laser would scan line by line. The software adjusts laser power according to image intensity, so dark areas are burned more while light areas use reduced power or are left as the original wood color.

Approximate settings for the test were as follows:

• Material: MDF sheet
• Engraving area: about 90 × 90 mm
• Laser head: 5W diode laser
• Wavelength: 445 nm
• Image mode: grayscale or dithered bitmap
• Toolpath type: raster engraving
• Maximum laser power: about 75–85%
• Speed: about 2,500–3,000 mm/min
• Line interval: about 0.10 mm
• Resolution: about 254 DPI
• Number of passes: 1
• Runtime: about 35–45 minutes

Before engraving, a laser test grid should be run on a scrap of the same material to compare power and speed and find values that produce a dark mark without burning too deeply. Different woods vary in color, density, glue, and moisture, so the same values should not be used on every sheet without testing first.

The machine reproduced the details of mountain lines, pine branches, water surfaces, and small house elements quite clearly. Dark-to-light shading helped give the image dimension. Even though the laser is only 5W, it has enough resolution for image work, signs, and small souvenirs.

During the test, we turned on the Makera Cyclone dust collector lite to help extract smoke and odor from the work area, because engraving MDF produces smoke from both wood fibers and the binders inside the board. After the job finishes, the extraction system should still be left running for a while before opening the machine cover.

If a darker image is needed, laser power can be increased or speed reduced, but that also increases the chance of burn marks and fuzzy edges. Reducing line interval makes details denser, but it greatly increases runtime. For an image like this, 0.10 mm offered a good balance between detail and time.

Makerables hub

This is a platform for collecting and sharing CNC projects. It includes files from Makera as well as community user work, covering 2D milling, 3D relief, laser engraving, and 4-axis milling. Users can view details, download files, or open them in Makera Studio to prepare and send jobs directly to the Makera Z1. Tool type, material, and cutting values should still be checked. It is very convenient.

The project detail page on Makerables shows sample images, creator information, CNC files, approximate runtime, and the number of tools used. Users can download the files or choose Open in Makera Studio to open the file, prepare the toolpath, and send the job to the Makera Z1 immediately. The example in the image is a wooden coaster project with a mandala pattern, which helps beginners try a job from a ready-made file without designing a model from scratch. Material size, bits, and cutting values should still be checked before real use.

Makera Z1 Desktop CNC video review

Conclusion

The Makera Z1 Desktop CNC machine is designed to be easier to use than a typical open-frame CNC machine. It uses a one-piece cast aluminum frame, a fully enclosed cover, a 200 × 200 × 100 mm work volume, and a 150W spindle with a maximum speed of 13,000 RPM. The standard model’s motion system uses linear rails with Acme lead screws, offering 0.02 mm motion resolution and spindle runout of about 0.01 mm.

The standout features include the quick tool change, auto probing & leveling, a wired probe, a built-in camera, and AeroDust, as well as the Makera Studio software that helps generate toolpaths without requiring every setting to be configured manually like on industrial CAM software. Even so, users still need to learn about bit types, spindle speed, feed rate, step-down, and workholding, because incorrect values can still break bits, ruin workpieces, or cause missed motor steps. This is especially true of the standard Z1, which uses open-loop motors and cannot detect lost position the way the Z1 Pro can.

In our review, the Makera Z1 could handle a wide range of work, from letter milling and small-detail engraving to 3D relief, 4-axis milling, and wood engraving with the 5W laser head. It can also mill aluminum and brass in theory, but only with suitable bits and cutting values and with depth per pass reduced to a level the machine can handle. This machine is better suited to small metal jobs and shallow milling than to large-volume metal removal. If metal milling is the main use case, continuous operation is required, or higher feed rates are needed, the Z1 Pro with ball screws and closed-loop stepper motors would be a better fit.

High-detail workpieces such as relief images and 4-axis sculptures can retain detail well, but note that such jobs can take a long time (many hours) because the software generates closely spaced toolpaths. The vacuum bed can conveniently hold flat sheet materials and keeps the top surface free of clamp obstructions, but air leaks must be controlled and cutting values that generate excessive side force should be avoided.

Overall, the Makera Z1 is a desktop CNC machine that significantly reduces workflow complexity, especially when setting zero, changing bits, generating toolpaths, and connecting accessories. It is not a machine that can be used straight out of the box with no learning at all, but once the basics of CNC are understood and settings are adjusted correctly for each material, it can produce detailed parts and handle a wide range of jobs in a compact footprint.

We would like to thank Makera for sending the Makera Z1 Desktop CNC and some accessories for review. The machine can be pre-ordered for $1,099. Accessories such as the 4th axis module, 5W laser module, vacuum bed, low-profile vise, and Cyclone Dust Collector Lite are sold separately or included in some packages.

CNXSoft: This article is a translation of the original review on CNX Software Thailand by Kajornsak Janjam, edited by Suthinee Kerdkaew

The post Makera Z1 Desktop CNC Review – Smart milling, 4-th axis module, and 5W laser engraving appeared first on CNX Software - Embedded Systems News.

ArmSoM Sige6 is yet another SBC powered by the Allwinner A733 octa-core SoC, offered in a credit card form factor with up to 128GB soldered-on eMMC flash and up to 16GB LPDDR5.

The single board computer features a microSD card slot and an M.2 M-key PCIe 3.0 x2 slot for storage, Gigabit Ethernet (with optional PoE), WiFi 6, and Bluetooth 5.4 connectivity, HDMI 2.0 and MIPI DSI display interfaces, a MIPI CSI camera interface, USB 3.0/2.0 ports, a 3.5mm audio jack, a 40-pin GPIO header, and more.

ArmSom Sige6 specifications:

• SoC – Allwinner A733-HN3
  • CPU
    • Dual-core Arm Cortex-A76 @ up to 2.00 GHz
    • Hexa-core Arm Cortex-A55 @ up to 1.80 GHz
    • Single-core RISC-V E902 real-time core
  • GPU – Imagination Technologies BXM-4-64 MC1 GPU with support for OpenGL ES 3.2, Vulkan 1.3, and OpenCL 3.0
  • VPU
    • 4Kp60/8Kp24 H.265/VP9/AVS2 decoding, 4Kp30 H.264  decoding
    • 4Kp30 H.265/H.264 encoding
  • AI accelerator – 3 TOPS NPU
  • Process – 12 nm
• System Memory – 2GB, 4GB, 6GB, 8GB, 12GB, and 16GB LPDDR5
• Storage
  • 32GB, 64GB, or 128GB eMMC flash
  • M.2 M-Key slot (PCIe 3.0 x2) for NVMe SSD
  • MicroSD card slot
  • 64 MBit SPI flash (option for 128Mbit or 256Mbit)
• Display interfaces
  • HDMI 2.0 output up to 4K at 60Hz
  • 4-lane MIPI DSI interface
• Camera interface – MIPI CSI camera connector; note: some of the specs imply there’s a second MIPI CSI connector, so maybe multiplexed with MIPI DSI?
• Audio
  • 3.5mm audio jack
  • HP-OUT connector for external speaker
  • Built-in microphone
• Networking
  • Gigabit Ethernet RJ45 port; PoE support via HAT
  • WiFi 6 and Bluetooth 5.4
• USB
  • 1x USB 3.1 host port up to 5 Gbps
  • 1x USB 2.0 host port
  • 1x USB 2.0 OTG Type-C port (also used for power)
• Debug – 3-pin UART header for debug
• Expansion
  • 40-pin Raspberry Pi-compatible header with GPIO, UART, SPI, I2C, PWM, 5V, 3.3V, GND
  • M.2 M-Key slot (PCIe 3.0 x2) for storage or AI accelerator
• Misc
  • Power, Boot, and Reset buttons
  • 2x LEDs (status and user)
  • 2-pin 5V fan header
  • Battery connector for LK8563S RTC chip
• Power Supply
  • 5V via USB-C connector
  • 5V via GPIO header (Pin 2/4)
  • Optional PoE module via 4-pin header
• Dimensions – 89 x 56 x 21.6 mm
• Weight – 47.2 grams
• Temperature Range – -25°C to +115°C (TBC, since the claim/range is unusual)

The company claims official support for Debian Linux and Android 13, and third-party support from Armbian (Debian/Ubuntu). However, right now, all I can find is a Debian 11 Gnome image, a closed-source csi_test_mplane sample for the camera, and Windows development tools like PhoenixSuit for Allwinner SoC flashing. More details can be found on the documentation website, although note that some information appears to be out of place (Allwinner T527 SDK with Linux 5.15). [Update: the company confirmed the T527 and A733 SDKs are the same]

The ArmSoM Sige6 is very similar to the Orange Pi 4 Pro and Radxa Cubie A7A, and the main difference is that ArmSoM offers soldered-on eMMC flash instead of pluggable eMMC/UFS modules, although the layout and some connectors (MIPI) are also different. ArmSoM and Banana Pi typically work together, but I could not find a Banana Pi equivalent for the Sige6. The upcoming Banana Pi BPI-M8 is also a credit card-sized Allwinner A733 SBC, but it has a slightly different layout.

The basic model of the Sige6 SBC with 6GB LPDDR5 and 32GB eMMC flash is up for pre-order for $119. The company told CNX Software that shipping is expected by mid-September.  For reference, the Orange Pi 4 Pro 6GB was introduced for $40 (not available now), and the Radxa Cuebie A7A 6GB sells for about $77, but that’s without a 32GB eMMC flash module, which adds $23 (if you can find it in stock). So the ArmSom’s price isn’t super low, but it’s not overly expensive either; it’s just the world we live in now…

The post ArmSoM Sige6 Allwinner A733 SBC features up to 128GB soldered-on eMMC flash, 16GB LPDDR5 appeared first on CNX Software - Embedded Systems News.

ARK Just A Pi is a tiny NDAA-compliant carrier board for the Raspberry Pi CM5 designed for drones, robots, or other space-constrained embedded applications that’s about the size of the Compute Module itself.

This follows ARK Electronics’ NDAA-compliant ARK Jetson Orin NX/Nano bundles introduced in 2024. The new “Just A Pi” is much smaller and lighter, features a built-in 100 Mbps Ethernet switch with two JST-GH Ethernet ports, a micro HDMI port, two MIPI CSI camera connectors, a microSD card slot, a PCIe FFC connector, a USB 3.0 Type-C dual role connector, and a range of small connectors for UART, SPI, PWM, GPIO, and fan. It also features a power connector for the company’s PAB power modules.

ARK Just A Pi specifications:

• Supported SoM
  • Raspberry Pi Compute Module 5 – Full support
  • Raspberry Pi Compute Module 4 – The following interfaces are not supported: CSI1/CAM1, UART2, USB-C, PCIe, and fan
• Storage – MicroSD card slot for CM4/CM5 Lite without eMMC only
• Video Output – Micro HDMI port
• Camera I/F – 2x 22-pin 0.5mm pitch FPC camera connectors compatible with Raspberry Pi 5 CSI pinout
• Networking
  • 2x 4-pin JST-GH connectors for dual 10/100 Mbps Ethernet
  • Built-in Ethernet switch
• USB
  • USB 3.0 Type-C dual-role port
  • USB 2.0 via 4-pin JST-GH connector
• Serial
  • 6-pin JST-GH connector for UART0 DEBUG/GPIO
  • 6-pin JST-GH connector for UART2/SPI3/GPIO
  • 6-pin JST-GH connector for UART3/SPI4/GPIO
  • 6-pin JST-GH connector for UART4/PWM/GPIO
• PCIe – 16-pin PCIe Gen2/Gen3 x1 FFC connector matching the Raspberry Pi 5 PCIe connector layout
• Misc
  • 4-pin JST-SH fan connector (same as on Raspberry Pi 5)
  • Built in the USA, NDAA compliant
• Power Supply – 5V/2A+ via 6-pin Molex Clik-Mate (also with I2C for ARK PAB power modules)
• Dimensions – 55 x 40 x 13.1 mm
• Weight – 19.5 grams

If you purchase a bundle with a CM5 and a microSD card, the system will come with ARK OS pre-installed, giving you access to the ARK UI dashboard for configuration. If not, I understand you’ll need to install Raspberry Pi OS 64-bit first, modify some parameters in /boot/firmware/config.txt, and download the latest ARK OS Pi Bookworm/Trixie package on GitHub to complete the installation. You’ll find (partial) instructions to get started and more technical details on the documentation website.

NDAA-compliant hardware is never cheap, and ARK Electronics sells the ARK Just A Pi carrier board for $335, or you can select a bundle with the board, a Raspberry Pi CM5 Lite 2GB Wireless, and a 64GB NDAA-compliant microSD card with ARK UI pre-installed for $475 on the same page. As a side note, if you’re just interested in getting a Compute Module-sized carrier board for the CM5 without optimization for drone/robotics use cases and/or NDAA compliance, the Waveshare CM5-NANO-B carrier board may serve your needs better at $17 to $28.

The post ARK Just A Pi – A tiny NDAA-compliant Raspberry Pi CM5 carrier board for drones and robots appeared first on CNX Software - Embedded Systems News.

D-Robotics RDK S100P development kit is a robotics SBC built around an S100P-based module providing an alternative to NVIDIA Jetson Orin NX 16GB with a 6-core Cortex-A78AE application cluster, a 4-core Cortex-R52+ MCU domain, a Nash BPU rated at 128 TOPS INT8, a Mali-G78AE GPU, and 24 GB of LPDDR5.

The four R52+ cores can be configured for reliable real-time control: two cores run in lockstep for safety, and the other two support split-lock operation. In practice, the A78AE CPU and BPU can handle Linux and vision processing, while the MCU handles time-critical tasks such as motor and sensor I/O. This can reduce the need for a separate real-time controller, although whether it eliminates one depends on the robot and its requirements.

RDK S100P specifications:

• SoC – D-Robotics S100P
  • CPU – 6x Arm Cortex-A78AE @ 2.0 GHz (safety-capable AE cores)
  • MCU – 4x Arm Cortex-R52+ @ 1.2 GHz (1x DCLS lockstep pair + 2x split-lock)
  • BPU – 1x Nash, 128 TOPS INT8 equivalent; vendor claims 160+ ONNX ops
  • GPU – Arm Mali-G78AE, 100 GFLOPS
• System memory – 24 GB LPDDR5, 96-bit bus, up to 6400 Mbps
• Storage
  • 64 GB eMMC flash (boot device)
  • M.2 M-Key (PCIe 3.0 x1) socket for NVMe SSD (official docs still say the board boots from eMMC only)
• Display – HDMI 1.4, up to 2560 x 1440 @ 60 Hz
• Camera
  • 100-pin expansion header
  • 3x 4-lane MIPI CSI-2
  • Optional camera board adds 2x MIPI + 4x GMSL (MAX96712)
• Audio – I2S/PCM on the 40-pin header
• Networking
  • 2x Gigabit Ethernet RJ45 ports (eth0 for general use, eth1 with built-in static IP 192.168.127.10)
  • Wi-Fi and Bluetooth via M.2 E-Key (PCIe 3.0 x1, UART) socket
• USB
  • 4x USB 3.0 Type-A ports
  • 1x USB 2.0 Type-C port (for flashing and debugging)
• Expansion
  • 40-pin main GPIO (SPI, I2C, I2S, PWM, UART)
  • 16-pin MCU GPIO
• Debug – JTAG header
• Misc – reset/sleep buttons, RTC battery connector, fan connector
• Power – 12-20V DC input (90W adapter included, board supports up to 150W max)
• Dimensions – 121 x 120 x 51 mm (including case)
• Operating temperature – 0°C to 45°C

The RDK S100P runs RDK OS, an Ubuntu 22.04 (Jammy)-based desktop image with Linux 6.1 (Current V4.0.3 notes list kernel 6.1.112-rt43 and U-Boot 2022.04, while newer downloads are around V4.0.4-beta). The OS can be flashed using D-Navigation from Windows, Linux, or Apple Silicon Macs through the Type-C port in U-Boot/fastboot or DFU mode.

For robotics development, the software stack includes TogetheROS.Bot, based on ROS 2 Humble, with components such as hobot_sensor for sensors, hobot_dnn for running AI models on the BPU, and hobot_codec for hardware video processing. Camera and multimedia functions are handled by hobot_vio and libsrcampy. Developers can also use the rdk_model_zoo, which includes models such as YOLO, and prepare them for the Nash BPU using the OpenExplorer toolchain and INT8 quantization. D-Robotics also provides the source code and tools needed to build custom system images, although some features, including BMI088 IMU support on the MCU expansion board, were still unavailable in V4.0.3.

D-Robotics also talks about two expansion boards for the S100 series. The RDK S100 Camera Expansion Board adds two 4-lane MIPI CSI-2 camera interfaces and four GMSL2 inputs ( via Fakra-Mini 4-in-1 Z-code connector) through a Maxim MAX96712 deserializer. The board can power GMSL cameras from the mainboard, but an external 12V supply is needed when their combined current exceeds 700mA. Supported cameras include the IMX219, SC230AI, SC132GS, SG8S-AR0820C, and Intel RealSense D457 and D435i.

The second expansion board, the RDK S100 MCU Port Expansion Board, connects to the S100’s Cortex-R52+ MCU domain and adds interfaces more commonly needed for robot control. It provides five CAN FD ports supporting up to 8 Mbps, a Gigabit Ethernet port, and a 30-pin header with up to seven ADC channels, two I2C interfaces, and two SPI interfaces. It also includes a Bosch BMI088 IMU, although the V4.0.2 SDK does not yet support it.

D-Robotics previously introduced the RDK X3 and RDK X5 development boards based on the company’s Sunrise X3 and X5 chips, with 5 TOPS and 10 TOPS AI performance, and the new RDK S100P is completely different and much more powerful beast.

Looking at the specifications, the RDK S100P could serve as an alternative to the NVIDIA Jetson Orin NX 16GB rather than the cheaper Orin Nano Super or the higher-end AGX Orin. Both use Cortex-A78AE CPUs, LPDDR5 memory, MIPI camera interfaces, and are designed for ROS 2 robotics. The Orin NX 16GB has 8x A78AE cores, a 1024-core Ampere GPU, 32 Tensor Cores, 16GB of 128-bit LPDDR5 memory, and up to 100 sparse INT8 TOPS at 25W. The S100P has 6x A78AE cores at 2.0 GHz, a Nash BPU rated at 128 INT8 TOPS, and 24GB of LPDDR5 on a 96-bit bus. These TOPS figures are not directly comparable, since NVIDIA and D-Robotics use different ways of measuring AI performance.

But the biggest difference is in the software and control hardware. Jetson offers CUDA, TensorRT, PyTorch, and Isaac, along with a large ecosystem of carrier boards and modules. The S100P uses D-Robotics’ OpenExplorer and .hbm model format, with TogetherROS.Bot is based on ROS 2 Humble. It also has a 4-core Cortex-R52+ MCU for real-time control, including lockstep and split-lock modes, which the Orin NX lacks. Power and form factor are also different, and the RDK S100 Module seems harder to source, at least for now.

The full RDK S100P development kit is sold for $899 on DFRobot or $999.99 on Waveshare. We initially thought we found a cheaper option at $625.99 on Amazon, but that’s the previous-generation RDK S100 (no P) with 12GB of RAM and delivering up to 80 TOPS instead.

The post D-Robotics RDK S100P – A 128 TOPS alternative to NVIDIA Jetson Orin NX 16GB with Cortex-A78AE/R52+ cores appeared first on CNX Software - Embedded Systems News.

Orange Pi Zero 4 is an affordable yet feature-rich Allwinner A733 single board computer (SBC) with up to 8GB RAM, eMMC/UFS storage footprint, a microSD card slot, and plenty of I/Os.

The board packs mini HDMI and USB-C DisplayPort, Gigabit Ethernet, WiFi 6 and Bluetooth 5.4, two MIPI CSI camera connectors, a Raspberry Pi-style PCIe FFC connector, and two GPIO headers into a compact 55x50mm form factor.

Orange Pi Zero 4 specifications:

• SoC – Allwinner A733
  • CPU
    • Dual-core Arm Cortex-A76 @ up to 2.00 GHz
    • Hexa-core Arm Cortex-A55 @ up to 1.79 GHz
    • Single-core RISC-V E902 real-time core up to 200 MHz
  • GPU – Imagination Technologies BXM-4-64 MC1 GPU with support for OpenGL ES 3.2, Vulkan 1.3, OpenCL 3.0
  • VPU
    • 8Kp24 H.265/VP9/AVS2 decoding
    • 4Kp30 H.265/H.264 encoding
  • AI accelerator – 3 TOPS NPU
• System Memory – 1GB, 2GB, 4GB, 6GB, or 8GB LPDDR5/LPDDR4/LPDDR4x; note: 16GB is also listed in the specs, but not for sale right now
• Storage
  • MicroSD card slot up to 128GB
  • Footprint for 8GB, 16GB, 32GB eMMC flash
  • Footprint for 32GB, 64GB, or 128GB UFS 3.0 module
• Video Output
  • Mini HDMI 2.0 port up to 4Kp60 resolution
  • USB-C port with DisplayPort Alt mode up to 4Kp60
  • Support for dual independent display setups
• Camera interface – 2x 4-lane MIPI CSI connectors
• Networking
  • Gigabit Ethernet RJ45 port via MotorComm YT8531C controller
  • Dual-band WiFi 6 and Bluetooth 5.4 module (Cdtech 20800D8-00) + IPEX antenna connector
• USB – USB 3.1 OTG Type-C port with DisplayPort 1.4 Alt. mode
• Expansion
  • 26-pin color-coded GPIO header with 17x GPIO, 4x UART, 3x I2C, 1x SPI, 5V, 3.3V, and GND
  • 13-pin “function interface” header with 8x GPIO, I2S, USB, IR Rx, 5V, and GND
  • 16-pin PCIe Gen3 x1 FPC connector
• Misc
  • Power and BOOT through holes
  • 2x LEDs
  • 2-pin fan header
• Power Supply – 5V/3A via second USB-C connector
• Dimensions – 55 x 50 mm
• Weight – 21 grams

Orange Pi plans to release Orange Pi OS (Arch), Ubuntu, Debian, and Android images for the Pi Zero 4, but as usual at launch, all links on the product page point to empty Google Drive shares. They should be uploaded soon, although looking at the Orange Pi Zero 3W status, I can only find Ubuntu Jammy/Resolute, Debian Bookworm/Bullseye/Trixie, and Android 13 images, but nothing for Orange Pi OS, so they probably dropped support. It is supported by Armbian with the vendor kernel, so we can expect the Orange Pi Zero 4 to get the same treatment. Early one there were talks about Allwinner being serious about mainlining A733, but nothing has happened, so I guess we can forget about it.

The Orange Pi Zero 4 is fairly similar to the earlier Orange Pi Zero 3W, except it adds Gigabit Ethernet, two LEDs, and switches the 40-pin GPIO header to two headers (26-pin + 13-pin). The form factor is also different: 55x50mm vs 65×30 mm. It’s also a more compact solution than the Orange Pi 4 Pro credit-card-sized SBC launched last week, and competes more directly with the Radxa Cubie A7S with a similar feature set and form factor (51 x 51 mm) except for having a MIPI DSI connector instead of mini HDMI.

The post Orange Pi Zero 4 – Compact Allwinner A733 SBC offers HDMI, USB-C DP, GbE, WiFI 6, PCIe FFC connector for $26.5 and up appeared first on CNX Software - Embedded Systems News.

Radxa Linkr is a compact, USB flash drive-sized IP KVM that connects to the target through HDMI and USB for video, keyboard, and mouse emulation. The user can control it over USB-C, Ethernet, or WiFi client/access point, even before the OS is booted.

The device does not require any specific apps, since all you need is a web browser. The company also highlights support for Tailscale VPN, two-factor authentication, and  AI agents such as OpenClaw and Hermes.

Radxa Linkr specifications:

• SoC – Rockchip RV1106G3
  • CPU – Arm Cortex A7 @ 1.2GHz with an integrated RISC-V co-processor.
  • NPU – 1 TOPS 4th-gen Rockchip NPU with hybrid quantization (int4, int8, int16)
  • ISP – 3rd-gen 5MP high-performance ISP
  • VPU – 3072×1728 (5MP) @ 30fps H.265/H.264 encoding, 16MP @ 60FPS JPEG snapshot
  • System Memory –  256MB DDR3L on-chip memory
• Storage for OS and virtual media images
  • Optional 16 GB eMMC flash
  • MicroSD card slot
• Display Input – HDMI up to 2K at 30 Hz
• Networking
  • USB-C network port (USB-C to RJ45 10/100 Mbps Ethernet adapter provided)
  • Wi-Fi 6 (AIC8800) with client and hotspot support
• USB
  • 1x USB-C port for keyboard and mouse emulation
  • 1x USB-C Network port for direct PC connector and RJ45 via adapter
• Misc
  • User Button
  • Video Status LED for HDMI input signal status
  • System Status LED for system status
• Power Supply – 5V/2A via USB-C port
• Dimensions – 70 x 22 x 12 mm

I’ve received a Linkr sample for review, so before checking out the software and documentation, let’s have a look at the hardware. I received the retail package for the Chinese market for the Radxa Linkr Model MV004.

The package contains the IP KVM device, a USB-C to RJ45 cable, a USB-Y cable for power and data, two USB-C to USB-A cables, and a bilingual (English/Chinese) quick start guide explaining how to connect the device to the target (USB-C, or network via Ethernet/WiFi) and access it through the Linkr.now website.

A teardown shows an AIC8800 wireless module for WiFi and a RealTek RTL8152B USB to Ethernet chip.

I initially thought the chip under the QR code would be an HDMI to CSI bridge, but the TLI TLiFE50AMSH3-MRE is a 16GB eMMC flash instead.

After taking the board out of the case, we can see the main chip is a Rockchip RV1106G3 SoC, not really a surprise since it’s used on other KVM over IP devices like the JetKVM, Luckfox PicoKVM, or AweSun Cloud KVM Q1. The HDMI-to-MIPI CSI bridge appears to be a Rockchip RK682F, a close cousin to the RK628D. A pink thermal pad is used to make sure everything runs cool.

I won’t have time to test it today, and for a while since I’ll be on the road, but you’ll find instructions to get started with the USB-C direct, USB-C to RJ45, WiFi client and access point methods, along with Tailscale, 2FA, and advanced usage (Wake-on-LAN, custom EDID, API access, and OpenClaw/AI agents integration) on the documentation website.

The Radxa Linkr is listed for $59 and up on AliExpress, Amazon, Arace, and Allnet, but is currently out of stock on all four stores. Additional information may be found on the product page.

The post Radxa Linkr – A USB drive-sized IP KVM with USB-C, Ethernet, and WiFi support appeared first on CNX Software - Embedded Systems News.

ACEMAGIC sent us a sample of the Kron Mini K5 Wildcat Lake mini PC for review. The computer is offered with either an Intel Core 3 304 or Core 5 320 CPU, and features triple display output, 2.5GbE and WiFi 6 networking, and several USB ports, but, interestingly, no Thunderbolt/USB4 ports.

The company provided us with the entry-level variant pairing an Intel Core 3 304 penta-core processor with 12GB of RAM and 512GB of storage. In this first part of the review, we’ll list the mini PC’s specifications, go through an unboxing and a teardown, before booting it into Windows 11. In the next parts, we’ll review it using Windows 11 and Ubuntu 26.04 in depth with feature testing and benchmarks.

ACEMAGIC Kron Mini K5 (Intel Core 3 304) specifications

The company hasn’t put up a product page for this exact model just yet, so we gathered specifications from our sample and information shared on the company’s social media accounts.

• SoC – Intel Core 3 304 (Wildcat Lake) processor
  • 5-core CPU – 1x P-cores @ 1.5/4.3 GHz (Turbo) + 4x LPE-cores @ 1.4/3.3 GHz (Turbo)
  • Cache – 6MB L3 cache
  • GPU – 1-core Intel Xe3 Graphics @ 2.3 GHz (9 TOPS)
  • NPU – 15 TOPS
  • Manufacturing process – Intel 18A
• Memory – 12 GB LPDDR5 5600MT/s (soldered on mainboard)
• Storage
  • 512 GB NVMe SSD in M.2 (PCIe 4.0 x2) slot, upgradeable up to 4TB
  • Additional M.2 2280 (PCIe 4.0 x2) slot, up to 4TB (could also be used by an AI accelerator)
• Video output
  • HDMI 2.0
  • DisplayPort 1.4
  • USB-C port with DP 1.4 Alt Mode
  • Supports triple displays up to 4K @ 60Hz
• Audio
  • 3.5 mm headphone/audio jack
  • Digital audio output via HDMI and DisplayPort
• Networking
  • 2.5GbE RJ45 port
  • Wi-Fi 6 + Bluetooth 5.2 (M.2 2230 module)
• USB
  • 2x USB 3.2 Gen2 Type-A ports up to 10Gbps
  • 4x USB 3.2 Gen1 Type-A ports up to 5Gbps
  • 1x USB-C 3.2 Gen2×2 port up to 20Gbps
• Misc
  • Power button
  • Kensington lock slot
  • Top cover lock/unlock switch
  • 80 × 10 mm cooling fan over 2,000 RPM with a large aluminum heatsink
• Power supply
  • 19V/3.42A via 5.5 × 2.1 mm DC jack
  • USB-C PD input up to 100W
• Dimensions – 128 x 128 x 41 mm, excluding rubber feet (aluminum chassis)

It’s a competitor to the Beelink EQi Wildcat Lake Core 3 304 mini PC with the same processor, a little less memory, different video output options, and no 10GbE or Thunderbolt 4/USB4 ports. The latter were found not to work well with several accessories (eGPU and NVMe SSD enclosure), at least at this stage, so the lack of USB4 ports might not be a big issue.

Unboxing

We received the sample in a retail package marked ACEMAGIC Kron Mini K5, but nowhere on the box is the CPU mentioned. However, the bottom indicates that our computer has 12GB of RAM and 512GB of storage.

The package includes the mini PC itself (with an Intel Core 3 sticker), a power adapter with power cord, an HDMI cable, a VESA mount and screw kit, and a multilingual user manual.

The power supply takes 100–240V AC @ 50/60Hz, 1.5A input, and outputs 19V DC up to 3.42A (64.98W)

The underside of the mini PC has the fan air intake and a sticker with the basic specs: ACEMAGIC Kron Mini K5, 12GB RAM, 512GB “ROM”, an Intel Core 3 304 CPU, and 19.0V ⎓ 3.42A DC input.

The front panel features one USB-C 3.2 port with DisplayPort Alt mode and USB PD, two USB 3.2 Gen2 (10 Gbps) Type-A ports, a 3.5 mm audio jack, and the power button.

The rear panel comes with HDMI 2.0 and DisplayPort 1.4 video outputs, a top cover lock/unlock switch, a 2.5GbE port, four USB 3.2 Gen1 (5 Gbps) Type-A ports, the DC jack, ventilation holes, and a Kensington lock slot.

The RJ45 jack is covered by a yellow label that reads “Kindly Note: For faster setup, avoid LAN initially. Connecting might trigger long updates, delay-ing desktop access”.  That’s the first time we’ve seen this warning. We found it funny and sad at the same time, since Windows 11 should have a better onboarding experience without manufacturers feeling the need to add such a warning.

The other two sides have ventilation openings but no extra ports.

ACEMAGIC Kron Mini K5 teardown

Let’s start the teardown by sliding the switch on the rear panel to the left to unlock, then slide the top cover.

We were expecting access to the NVMe SSD and expansion sockets at this point, but we had to loosen a few more screws first to remove the cooling plate.

We can finally check out the mainboard. It’s connected to a CMOS battery and features three M.2 sockets, two of which are used by a COLORFIRE CF620 512GB M.2 2280 NVMe SSD (PCIe Gen3 x4, MAP1202 controller) and a Cdtech CDW-C9852BE-00 PCIe + USB WiFi 6 and Bluetooth 5.2 module based on a Realtek RTL8852BE chip. The remaining M.2 socket (H1_SSD2) can be used for another 2280 NVMe SSD or AI accelerator.

First boot to Windows

Time to reassemble everything and try the mini PC. We connected it to a wireless keyboard and mouse combo and a USB-C display, and powered it using the provided power adapter through the DC jack. We went through the initial Windows 11 Pro setup wizard, setting up the language, Wi-Fi network, and so on, and eventually reached wth the Windows 11 Pro desktop with a working internet connection.

The System->About window confirms we have a 1.5 GHz (base frequency) Intel Core 3 304 processor, 12GB of RAM, and 477 GB of storage loaded with Windows 11 Pro 64-bit version 25H2.

That will be all for today. In the next part of the review, we will test the Kron Mini K5 (Intel Core 3 304) mini PC with Windows 11 in more detail with feature testing, system and AI benchmarks, power consumption measurements, and more. After that, we’ll install Ubuntu 26.04 to see how the Wildcat Lake mini PC performs on Linux.

We’d like to thank ACEMAGIC for sending the Kron Mini K5 (Intel Core 3 304) mini PC for review. At this point, we’d usually mention the price and where to buy the reviewed model, but the company has yet to sell the Core 3 304 and Core 5 320 variants. As an educated guess, we’d assume it to go for around $400 or less, since the Beelink EQi Wildcat Lake with 16GB RAM, 512GB SSD, 10GbE, and Thunderbolt 4 ports now goes for $469, and the BOSGAME ECO with a similar 12GB/512GB configuration, but a slightly different port mix, sells for $380.

Continue reading:

1. ACEMAGIC Kron Mini K5 Review – Part 2: A $400 Intel Core 3 304 Wildcat Lake mini PC tested with Windows 11 Pro
2. ACEMAGIC Kron Mini K5 Review – Part 3: Ubuntu 26.04 on an entry-level Intel Core 3 304 mini PC (Coming soon)

CNXSoft: This article is a translation – with some additional insights – of the original review on CNX Software Thailand by Suthinee Kerdkaew.

The post ACEMAGIC Kron Mini K5 (Core 3 304) review – Part 1: Specifications, unboxing, teardown, and first boot appeared first on CNX Software - Embedded Systems News.

IBASE AA400-N is an “ultra-compact” fanless edge AI computer powered by an AMD Ryzen Embedded 8840U/8640U processor with up to 8 cores, 16 threads, and a combined 39 TOPS of AI performance for machine and robotic vision, autonomous inspection, smart retail, industrial automation, and other data-intensive, space-constrained applications.

The most intriguing part of the specs and announcement is the claim of up to 256GB DDR5 capacity through two SO-DIMM modules, implying availability of 128GB DDR5 SO-DIMM memory modules. The system also comes with a 256GB M.2 NVMe SSD, features 8K-capable HDMI and DisplayPort video interfaces, two 2.5GbE jacks, M.2 sockets for optional WiFi, Bluetooth and 4G/5G cellular connectivity, a few USB ports and a COM port.

IBASE AA400-N specifications:

• AMD Ryzen Embedded 8000 Series processors
  • AA400-NM  – AMD Ryzen Embedded 8840U
    • CPU – 8-core/16-thread Zen4 “Hawk Point” processor up to 3.3 GHz / 5.1 GHz (Turbo)
    • GPU – AMD Radeon 780M with 12x Compute Units @ 2700 MHz (RDNA3 graphics)
    • AI
      • 16 TOPS NPU (AMD XDNA)
      • Total combined 39 TOPS of AI performance
    • TDP – 15W
  • AA400-N8640M –  AMD Ryzen Embedded 8640U
    • CPU – 6-core/12-thread Zen4 “Hawk Point” processor up to 3.5 GHz / 4.9 GHz (Turbo)
    • GPU – AMD Radeon 780M with 8x Compute Units @ 2600 MHz (RDNA3 graphics)
    • AI
      • 16 TOPS NPU (AMD XDNA)
      • Total combined ~31 TOPS of AI performance
    • TDP – 15W
• Memory – Up to 256GB dual-channel DDR5 5600 MHz via 2x 262-pin SO-DIMM sockets (2x 128GB)
• Storage –  256GB NVMe SSD via M.2 Key-M 2280 (PCIe Gen4 x4) socket
• Video output
  • HDMI 2.1 port up to 8K
  • DisplayPort 2.1 up to 8K
• Networking
  • 2.5GbE RJ45 port
  • Optional WiFi and Bluetooth via M.2 socket
  • Optional 4G LTE/5G cellular via M.2 socket and Nano SIM card slot
• USB
  • 2x USB 3.2 Gen1 Type-A ports
  • 4x USB 2.0 Type-A ports
• Serial – 1x COM port
• Expansion
  • M.2 M-Key-M (PCIe Gen4 x4) 2280 socket for NVMe SSD storage
    M.2 E-Key 2230 socket for WiFi and Bluetooth (CNVi)
  • M.2 B-Key 3052 socket for 5G/LTE
• Misc
  • Mounting Type – DIN rail, wall mount
  • Enclosure made of zinc-coated cold-rolled steel (SGCC)
• Power Supply – 12V ~ 24V DC via 4-pin DC connector; up to 45W
• Dimensions – 215.6 x 177.6 x 55.2mm
• Weight – 3 kg
• Temperature Range – Operating 0°C ~ 45°C; storage: -20°C~80°C
• Relative Humidity – 0%~ 90% (non-condensing)
• Vibration – 1 grms / 3~500Hz / random operation
• Shock – Operating: 20 g / 11 ms; non-operating: 40 g / 11 ms
• Certifications – CE, FCC, LVD

IBASE claims support for Ubuntu 22.04 and Windows 11 for the embedded system. The closest competitor is probably the SolidRun Bedrock R8000 industrial PC introduced in 2024, which at the time claimed up to 96GB of RAM through two SO-DIMM sockets (2x 48GB).

Let’s have a closer look at the 256GB DDR5 capacity claim using only two SO-DIMM sockets, which would require two 128GB DDR5 SO-DIMM modules. My initial research pointed to higher-capacity DIMM modules available for 128GB or more, but that would not work in a SO-DIMM socket, or the system would be too small to have two DIMM sockets. After a few minutes, I assumed the specs and the press release might have been wrong, but I eventually found the Centron S3-D5SUN2CH-128G.G8 DDR5 SODIMM unbuffered non-ECC module. So 2x 128GB DDR5 SO-DIMM is indeed possible if you have the cash and are a system builder.

IBASE hasn’t provided availability and pricing information, with the company only saying it “debuts the AA400-N”. Additional information can be found on the product page.

The post Fanless AMD Ryzen Embedded 8840U/8640U Edge AI Computer supports up to 256GB DDR5 SO-DIMM memory appeared first on CNX Software - Embedded Systems News.

Spark Solder is a compact, USB-C-powered one-handed soldering iron with a built-in motorized solder feeder for convenient one-handed operation. It supports standard TS100 tips, delivers up to 45W of heating power, and features a 1-inch LCD for adjusting the temperature and solder feed speed.

While smart, portable soldering irons are very common these days, using one can still be tricky when you have to handle the iron, solder wire, and components at the same time. Spark Solder solves this by adding a replaceable solder capsule and a two-axis motorized feeding system to the handle, leaving one hand free for easier assembly, prototyping, and repairs.

Spark Solder specifications:

• Display – 1-inch TFT display for adjusting temperature, solder extrusion speed, and tip profiles
• Soldering
  • Power output – Up to 45W
  • Temperature Range – 200°C to 450°C
  • Heat-Up Time – Reaches 300°C in 6 seconds
  • Temperature Stability – ±2°C (via PID control)
  • Tip Compatibility – Standard TS100-series tips
  • Feeder – Integrated solder capsule (comes with 40g of lead-free solder)
  • Precision two-axis solder delivery with a high-torque motor and spring-ball mechanism
• Safety – Overheat and power-safe protection
• Power supply – 20V via USB-C port
• Dimensions – 240 x 25 x 25 mm
• Weight – 100 grams
• Enclosure – Durable plastic body

The main feature of Spark Solder is its two-axis motorized solder delivery system, with which the solder wire can slide and rotate while a spring-ball mechanism keeps it securely in place to help prevent slips and jams. Users can also adjust the solder feed speed through the interface, allowing the wire to flow smoothly only when needed.

Previously, we wrote about other portable USB-C soldering irons such as the Miniware TS100, Pine64 Pinecil, and Pinecil V2, and the Spark Solder iron borrows the TS100 tip and USB-C concept from those devices, but adds an automatic solder wire feeder. What we were not able to find is information about the MCU and the firmware doesn’t seem to be open-source.

The one-handed, self-feeding concept itself is not new. For example, an article from the January 1972 issue of Popular Electronics showcased the Bilectro “One-Hander”, a pistol-grip soldering tool that fed solder when the trigger was pulled. More recently, affordable auto-feed soldering guns like the Saker have also appeared online. But these tools are generally quite bulky, use AC power, and rely on manual triggers that can be tiring and less precise. Spark Solder updates the idea with a lightweight 100-gram pen-style design, motorized solder feeding, digital PID temperature control, and USB-C PD power.

The Spark Solder one-handed soldering iron is available on Kickstarter, starting at a $64.99 Super Early Bird pledge (15% off the planned $79.99 retail price). A regular Early Bird tier is also available for $69.99, along with multi-packs including a 2-pack for $124.99 and a 4-pack for $224.99.

The kit includes the Spark soldering iron, a 40g solder capsule, one TS100 tip, a 45W USB-C wall charger, a high-current USB-C cable, and a user manual. According to the campaign timeline, mass production is scheduled for November 2026, with global backer shipping slated to begin in December 2026. The campaign page explicitly mentions that they used ChatGPT for campaign images and graphics, and MiniMax Audio for the video voiceover; however, the video demo below appears to be using a real prototype.

The post Spark Solder one-handed USB-C soldering iron supports TS100 tips, integrates solder feeder (Crowdfunding) appeared first on CNX Software - Embedded Systems News.

Pollen Robotics and Hugging Face have unveiled the Microduck, a 25 cm tall biped robot with 15 motors, a camera, a depth sensor, two motion sensors, and an articulated beak that can pick up objects.

This follows the launch of the Reachy Mini desktop robot earlier this year that featured a Raspberry Pi CM4 module or could interface to a computer instead. In some ways, we may think of the Microduck as a Reachy Mini with legs, but the company says both are different products. The Reachy Mini targets human-robot interaction, while the Microduck focuses on motion/action, although it also features a microphone and a speaker for audio interaction.

Microduck specifications:

• SoC – Rockchip RK3566
  • CPU – Quad-core Cortex-A55 processor @ up to 1.8 GHz
  • GPU – Arm Mali-G52
  • NPU – 0.8 TOPS AI accelerator
  • VPU
    • 4Kp60 H.265/H.264/VP9 video decoder
    • 1080p60 H.264/H.265 video encoder
• System Memory – 1GB RAM
• Storage – 32GB eMMC flash
• Camera – Front camera with a dedicated camera-use indicator
• Audio – Microphones and speaker, with a per-robot generated voice
• Wireless
  • Wi-Fi and Bluetooth
  • NFC with 2 antennas, one in the head and one in the beak
•  Sensors
  • 2x IMUs, one in the body and one in the head
  • Compact 8×8 ToF LiDAR (maybe something like VL53L8CX;  we’re also told range is not finalized)
• Motors: 15 degrees of freedom, across articulated legs, head, and neck
• Physical interaction – Articulated grasping beak
• Power Supply – 2600 mAh NP-F550 camera battery, good for around one hour
• Dimensions – It’s complicated, but: 25 cm tall, 14 cm wide
• Weight – < 800 grams

The robot will ship with a game controller, a battery, and a USB cable. It will be playable before writing any code, with seven trained moves at launch.

The open-source Rust software and SDK are hosted on GitHub. On the robot, that stack runs as daemons on the Rockchip RK3566 processor: a 50 Hz control loop for the fifteen servos, plus the radios, camera, depth sensor, and update engine. Simulation and RL training tools (Python code) are in a separate repo.

A Microduck simulator is hosted on Hugging Face to play around before getting the hardware. The companies have probably never heard of Firefox, as both the project’s website and simulator were unusable (as in really slow, Raspberry Pi 1 Model B style), and I had to switch to Chrome to kick a ball with my virtual duck. Instructions are for AZERTY keyboard users, but the simulator also works fine on a QWERTY keyboard using “AWSD” rather than the awkward “ZQSD” combination.

While the Microduck’s software stack is open source, nothing was said about the mechanical and hardware design files, and Pollen Robotics told the press not to refer to the robot as “open-source hardware” (for now).

Pre-orders are open on Pollen Robotics for $399 (use Chrome rather than Firefox), and the Microduck robot is only available in North America and Europe for now. Shipping is scheduled to start before Christmas 2026. A few more details may be found on the product page.

The post Microduck – A duck-like biped robot designed for physical AI experimentation and fun appeared first on CNX Software - Embedded Systems News.

The USB-C Deskradar is an ESP32-C3-based, aviation-themed ADS-B flight tracker that tracks and displays live aircraft positions around your location on a sonar-style circular grid. A 1.28-inch round display and a 3D-printed enclosure provide a real-time view of flights around a configured location.

As we’ll see further below, it appears to be an open-source project that was picked up by Chinese vendors and started popping up on AliExpress as an inexpensive desk gadget.

USB-C Deskradar specifications:

• Wireless module –16-pin ESP32-C3 Super Mini module
  • SoC – Espressif Systems ESP32-C3FH4, single-core 32-bit RISC-V microcontroller @ up to 160 MHz; 400 KB SRAM; 4MB flash
  • USB – USB Type-C port for programming
  • Ceramic antenna
• Display – 1.28-inch round TFT color LCD with GC9A01 controller; 240×240 resolution; driven via SPI
• Misc
  • Hardware button (connected to GPIO 9 / BOOT pin, active LOW)
  • Short tap – Cycles radar range presets (e.g., 5, 10, 15, 25 km)
  • Long press (3 seconds) – Wipes saved Wi-Fi/location credentials and reboots to setup mode
• Power Supply – 5V via USB Type-C
• Enclosure – Custom 3D-printed housing designed for desktop use (STL files provided)

The software handles flight tracking, radar display, and device setup, making the device easy to use and configure, with features designed to provide a simple way to monitor nearby aircraft. Key features include:

• Live ADS-B flight data from adsb.fi or OpenSky Network
• A sonar-style radar showing aircraft heading, speed, callsign, altitude, and type
• Wi-Fi Manager with a captive portal for Wi-Fi, location, and km/mi settings
• Runway overlays for major airports
• Arduino/PlatformIO firmware using LovyanGFX and ArduinoJson

I first learned about this project through an AliExpress listing, but there was barely any documentation or information on how it works. After searching around a bit, I found two open-source GitHub projects that appear to match the hardware and functionality: arvis91/deskradar and MatixYo/ESP32-Plane-Radar. The AliExpress device may be based on one of these projects, although I couldn’t find anything confirming which firmware it uses.

Once connected to the internet, the ESP32 regularly fetches live flight data from open APIs. The MatixYo firmware updates data from adsb.fi every 5 seconds, while arvis91 uses the OpenSky Network API. The 240×240 display shows a dark blue sonar-style radar with range rings, compass directions, aircraft symbols, speed vectors, and callsigns or ICAO codes. Aircraft outside the radar range are shown as directional dots around the edge.

As mentioned earlier, the device is available on AliExpress for $26.93, but if you prefer a DIY approach, you can build one yourself for less than $10. All you need is an ESP32-C3 Super Mini and a 1.28-inch GC9A01 round display, or a development board like the ESP32-2424S012. Alternatively, you can modify the firmware to add ESP32 support and use displays like Makerfabs ESP32-S3 Round SPI TFT with Touch 1.28 or Waveshare ESP32-S3-LCD-1.28, which includes an onboard ESP32-S3. The 3D-printable case files are available for free on MakerWorld, and once everything is assembled, you can flash the ESP32-Plane-Radar or deskradar firmware directly from your browser using esptool-js.

The post USB-C Deskradar – An ESP32-C3 ADS-B flight tracker with a 1.28-inch round display appeared first on CNX Software - Embedded Systems News.

Opener is an open-source reference implementation of the DECT NR+ protocol stack, formally known as DECT-2020 NR, used for massive IoT (mMTC) and low-latency (URLLC) machine-to-machine communication applications.

It is developed by the Opener Initiative, supported by the OpenDECT-X project, and funded by the German Federal Ministry of Research, Technology and Space (BMFTR).

The standard itself supports up to one million devices per km², years of battery life for low-power sensors,  non-IP and IPv6 traffic, mesh networking, and random-access operation. It is also designed for ultra-reliable low-latency communications (URLLC) with sub-millisecond air-interface latency, a 99.999% reliability target, and scheduled access for bounded latency.

Contrary to 3GPP cellular communication networks, DECT NR+ does not rely on centralized infrastructure and uses context-based rather than fixed device roles. It operates in the license-exempt spectrum in the 1880–1930 MHz range (1.9 GHz), and will be extended to the 3.8–4.2 GHz band, at least in Europe, following the EU’s Decision 2025/2425. Supported topologies include point-to-point links, star networks, and mesh.

Opener implements the MAC (Media Access Control) and upper layers of the NR+ protocol stack up to Application Profiles. It’s written in C, built for Zephyr RTOS, and released under the Apache 2.0 license for both research and commercial use. You’ll find the Opener implementation and code samples in the openener and opener-samples repositories on the Opener Initiative’s GitHub account.

While the project is designed to be vendor- and application-agnostic, you need to start somewhere, and Nordic Semiconductor’s nRF91 series is currently the only off-the-shelf integrated hardware for DECT-2020, and Dresden, Germany-based Last Mile Semiconductor will soon enter the DECT NR+ market.  Looking at the code sample repository, the Opener dectnrp driver specifically relies on the Nordic Semi nRF9151 DK, but should work on the nRF9151 SMA DK (same as the DK but with SMA connectors) and third-party nRF9151 boards like the Makerdiary nRF9151 Connect Kit, Norik Systems DECT NR+ USB dongle, Conexio Stratus Pro nRF9151 IoT development kit, and so on.

A few more details may also be found on the Opener Initiative website.

The post Opener is an open-source reference implementation of the DECT NR+ massive IoT, low-latency standard appeared first on CNX Software - Embedded Systems News.

We’ve gotten a sample of the LUNYEE 3020 Nova desktop CNC router for review. The machine features a 710W trim router spindle suitable for carving wood, acrylic, and aluminum. The drive system relies on dual HG15 linear guides and 1204 ball screws for enhanced rigidity and motion accuracy, and the machine integrates NEMA 17 stepper motors with 0.89 N·m torque. It supports Windows, Linux, and Mac systems and works with GRBL-compatible CNC software such as UGS, Fusion 360, VCarve, Easel, and LightBurn.

LUNYEE sent us the CNC router with a dust shoe and a CNC vise worktable for review. We will cover the specifications, go through an unboxing, and show how to assemble and set up the machine before testing it with UGS and Easel programs using wood and acrylic as test materials suitable for the bits provided by the company.

LUNYEE 3020 Nova desktop CNC router specifications

• Control Board – 32-bit 4-axis GRBL firmware-compatible controller
• Stepper Motor – NEMA 17 (60 mm size, 2.3A current, 0.89 N·m torque)
• Working Area – 300 x 200 x 80 mm
• Accuracy – ≤ ±0.05 mm
• Maximum Working Speed – 5,000 mm/min
• Spindle Power – 710W trim router with 6-speed adjustment
• Spindle Speed – 6,000 – 30,000 RPM
• Drive System – Dual HG15 Linear Guides and 1204 ball screws
• Safety Features – Limit Switches for X/Y/Z axes and emergency stop switch
• Power Supply – 48V DC up to 10A
• Dimensions – 52 x 41 x 39.3 cm
• Product Weight – 19.2 kg
• Frame Material – Full aluminum structure
• Supported Software – GRBL-compatible firmware such as Easel, Candle, UGS, Fusion 360, LaserGRBL, LightBurn

Compared to the LUNYEE 3018 Pro Ultra CNC Router 500W reviewed last year, the LUNYEE 3020 Nova ships with a higher power 710W spindle with six manually adjustable speeds from 6,000–30,000 RPM, providing a wider speed range than the 3018 Pro Ultra rated up to 12,000 RPM. The Y-axis working area is expanded from 180 to 200 mm, the drive system is upgraded from lead screw to 1204 ball screw, while dual HG15 linear guides are used on all 3 axes. Additionally, the NEMA 17 motor torque is increased from 0.55 to 0.89 N·m, making the machine more rigid and accurate, at the cost of increased weight from 14 to 19.2 kg.

Unboxing

Let’s start the review with an unboxing of the LUNYEE 3020 Nova.

The CNC machine comes almost fully assembled with the STOP (emergency) button attached to it, a 65 mm spindle mounting bracket with screws, a 710W trim router, a CNC vise worktable, manual and installation documents, an AC power cable, a USB data cable, an accessory box, ER11 1/8-inch collet set with bits, a USB flash drive, lubricating grease, one flat-head screwdriver and one wrench set, a cleaning brush, a 15 cm steel ruler, two workpiece clamps, cable ties, and a dust shoe. Spoiler: we now consider the latter a must when working with wood.

LUNYEE 3020 Nova assembly

While the LUNYEE 3018 Pro Ultra CNC router required some work for the assembly and cabling was sort of a mess, the 3020 Nova model comes mostly preassembled and wired out of the box, so users can get started in a few minutes.  We still had to install and secure the Z-axis bracket by tightening the hex screws.

After that, we inserted and secured the spindle mounting bracket to it before installing the trim router. We initially placed it roughly in the middle as per the photos in the user manual, but what you’d really want to do is lower the Z-axis to the maximum and adjust the Z-position of the trim router so that the end mill reaches the worktable in its lowest position, or even goes slightly lower than the worktable level and dips in the T-slots for straight cuts without a wasteboard.

One interesting note: the M1P-3701 trimmer is rated 220V, 50Hz, 800W, Ø6mm, with a maximum speed of 30,000 RPM, rather than the 710W advertised by the company.  The trimmer has a separate power cable and is not electrically connected to the controller, so it’s not controlled by the software, but manually by the user.

The next step is to connect the Z-axis motor to the control board with the black cable as shown below.

The dust shoe is an optional accessory we requested with the unit, since wood carving typically leaves dust everywhere, and it’s not fun to clean it up afterwards, and not ideal for health reasons.

Insert the dust shoe into the trimmer and secure it with the green lock.

Assemble the brush with the two magnetic rings, then install the brush set by aligning it with the magnets. We’ll remove these each time we adjust the X, Y, Z position before starting a job, and reinstall them before starting it.

Connect the dust extraction hose to your vacuum cleaner.

The whole assembly should take around 5 minutes or even less.

Preparing and setting up the CNC router

The LUNYEE 3020 Nova has a working area of 300 x 200 mm and an 80 mm Z-axis travel

The workplate features T-slots that we will use to install clamps to keep the workpiece (or the vise) secure.

The kit only includes 5 bits: 2x pointed tips and 3x 1/8-inch spiral flat nose end mills (3.175 mm). These work well for wood and acrylic, but are not ideal for aluminum.

To install an end mill, insert the ER11 1/8-inch collet into the nut, then insert the end mill into the collet and loosely screw it onto the spindle.

Finally, properly install the end mill and tighten it using two wrenches.

After that, connect the power adapter and check the emergency stop switch. Good news: it works fine, contrary to our experience with the LUNYEE 3018 Pro and 4040 PRO CNC routers… It has a larger design and immediately stops the machine when we press it.

We can now turn on the machine by pressing the power switch to get started.

Testing the LUNYEE 3020 Nova CNC router with the UGS Program

We’ll start testing with the UGS program as recommended in the wiki. UGS is an open-source, cross-platform G-Code sender for GRBL that works on Windows, Linux (x86 and Arm/Raspberry Pi), and macOS. A Windows laptop was used for this review, so we started by installing the driver (CH341SER.EXE) from the USB Flash Drive before installing the program, but you can also download it from the official CH340 website.

We will test the machine with one of the provided sample files using a 1/8-inch spiral flat nose end mill we’ve just installed. Connect the CNC machine to your computer via a USB cable, then open the UGS program, open an .NC file, and in the top bar, set Firmware to GRBL, select the COM Port, and set the baud rate to 115200.

It’s also possible to access the machine from your smartphone using a local web dashboard

If the machine status changes to ALARM when clicking on the connect icon, click Unlock to change the status to IDLE. Then open the G-code file, and check the sample G-code toolpath in the Visualizer window on the right.

Then go to the Jog Controller to move the bit to the starting point. The correct gap between the end mill and the workpiece is about the thickness of a sheet of paper: the paper should slide under the bit with slight resistance.

Click Reset Zero in the Toolbox window to set the zero point.

Alternatively, you could make use of the Z-Probe to calibrate the Z-axis.

Select Window -> Plugins -> Probe Module

Click Z, measure the Z-Probe thickness with a caliper, and enter the value in the Touch plate thickness field, then enter -10 mm in the Probe Distance/Direction field or another suitable value to indicate how far the Z-axis can move down. Click the “Probe and zero Z” button to complete the process.  We used the paper method during this review as we found it more convenient.

After placing the workpiece in the work area and securing it with two clamps, we installed the magnetic parts of the dust shoe, turned the spindle on (speed 3), turned the vacuum cleaner on, and clicked the Send/Start icon to start engraving.

The job completed successfully, and the dust shoe and vacuum cleaner did their job effectively with practically no dust on the machine or around. The only downside is that we couldn’t clearly see the job in action.

Nevertheless, the engraving came out beautifully.

Using the Easel Program for custom designs

The UGS program only sends G-code to the machine, so we need to use another program to create our own designs or parts. We went with Easel, created an account, and selected the free 30-day trial. It’s a web-based interface, so it works on Windows, Linux, and macOS.

We will test engraving the text “CNX” on a 62 x 32 x 5 mm plywood sample provided with the kit. Instead of using clamps on this small sample, we secured it with the CNC vise worktable, itself secured to the T-slots of the workplate with two screws.

Back to the Easel program, we set the dimensions of the workpiece, designed a “CNX” logo with the text tool, selected the material type (Birch Plywood), bit type ( 1/8-inch Spiral Flat Nose End Mill), and set the engraving depth to 3 mm. After creating it and running the simulation, it was time to click “Carve” to connect to the machine.

Select the USB Serial (COM6) port and click Connect. Note that we found that part to be fairly unreliable in Easel, so we had to either disconnect and reconnect the USB cable or/and connect and disconnect the machine in UGS before it worked in Easel. The root cause is unclear, but basically, if the Carve button is blue, that means the machine is not detected, and if it turns green, that’s a very good sign!

Once the connection is successful, we go through the Easel software wizard to confirm the material, check whether it is securely clamped, adjust the zero position with the Jog Machine, turn the spindle on (speed 3), turn on the vacuum cleaner, then click “Carve!” to start the job. While the machine is working, closely monitor it, listen for abnormal sounds or strong vibrations, and press the emergency stop button if any problem occurs.

This small job took about 3 minutes to complete. The result is not too bad, although we had to use some sandpaper to clean it up.

Next up, we cut the large wood plank in our very first test to get a smaller 116 x 114 workpiece for our next design. We cut it by having the end mill work its way in one of the T-slots and rotating the wood plank for the second pass. We then loaded a Linux Tux design from the Design Library in the Easel program using the same parameters, except for the engraving depth, which was set to 4 mm.

The size was perfect to use with the CNC vise worktable, and again we used the dust shoe for a clean job.

This design took about 11 minutes to complete.

The cut left some fuzzy edges and raised fibers, which we removed with sandpaper, and the final result looks much better.

We also tried to adjust the feed rate and the spindle rotation speed, but it didn’t change the results much.

Next, we will try using a different type of wood. We were unsure of the type, so we asked ChatGPT, which identified it as Burl wood (burl/eye wood). It is not available in the program, and the closest type is “Walnut” in Easel. We loaded a horse sample from the Design Library.

We set the engraving depth to 5 mm, but due to the uneven material thickness, the engraving result looked as shown in the image.

Normally, we’d flatten the piece first with a large end mill, but we didn’t have any, so it would have taken quite a long time. So instead, we used a spirit level to adjust it, but that was obviously not good enough.  It’s still kind of “artsy”.

Let’s now switch to a 200 x 200 x 3 mm acrylic plate with an engraving depth set to 1 mm using a Linux Tux design.

It took nearly 10 minutes, and the result came out beautifully.

Someone requested that we make a teak wood sign from a material around 790 mm long. The workpiece has uneven width (maximum 190 mm) and uneven thickness (maximum 50 mm). The goal was to cut Thai text “สวนสองสี่หก” (Garden Two Four Six) with a character height of 80 mm, width of 50 mm, and engraving depth of 10 mm.

However, Thai is a funny (complex) language from a computer perspective with vowels on the left, right, bottom, and top, and potentially tone marks above. The first time, we selected a random font that rendered the Thai text properly on the left side of the program, but showed rectangles (like characters not found) on the simulation side on the right. We didn’t pay attention, and after a while we noticed the rectangle on the workpiece; at which point we stopped the job. The lesson here is to always run the simulation before starting a carving job.

We fixed the Thai text somewhat by loading Thai fonts. However, there was still an issue. One of the consonants came with a vowel and a tone mark (mai ek) on top, and while the text shows correctly on the left side of the program, the tone mark was missing on the right side. Moving or resizing the text didn’t help, so we suspect the program cannot render a font with two glyphs on top of a vowel. One solution is to design the text in another program like Inkscape, export it as SVG, and import it in Easel. However, that’s a premium feature. Instead, it just cut the tone mark later after the main job was complete.

A 790 mm piece (and 600 mm artwork) is too long for a 3020 CNC machine, so we enabled tiling in Easel to divide the job into three parts, each with three characters.

Due to the uneven wood thickness, we used a spirit level and various supports to keep it horizontal as much as possible. We also used a pencil to mark the orientation of the workpiece on the workplate.

Each section took a little over 2 hours. Teak is hardwood, and the Z-axis was set to go down 1 mm at a time. The dust shoe was also really helpful for this job. Just make sure your vacuum cleaner is designed to work for several hours at a time without overheating. We did turn off the vacuum cleaner for a while during the first section, which explains why some dust shows up in the photo below.

After each section, we moved the wooden plank,  made sure it was parallel to the workplate as much as possible, and oriented it properly before securing it with two clamps. The job took around 7 hours in total.

Conclusion

The LUNYEE 3020 Nova is a capable CNC Router with a 300 × 200 × 80 mm working area and a powerful 710W trim router suitable for various types of wood, acrylic, and even aluminum. We tested it with UGS and Easel and could engrave wood, plywood, acrylic, as well as harder and larger teak wood.

Compared to the earlier 3018 Pro model, the LUNYEE 3020 Nova comes mostly pre-assembled and ready to be used within minutes, and it features a larger emergency stop button (that actually works). We highly recommend getting the dust shoe, especially if you plan to work with wood. By default, the machine comes with 5 small end mills that are fine with wood and acrylic, but not really designed for aluminum (except small, superficial jobs), so you’d need to bring your own, or add an end mill kit to your order.

On the software side, we had no issue with UGS, but it’s limited to sending G-code files. For designing projects, Easel is great and easy to use, but we experienced reliability issues when connecting to the machine. That’s only for the initial step when clicking the Carve button, and it doesn’t happen once the job is in progress.

We’d like to thank LUNYEE for sending the 3020 Nova CNC router for review along with a dust shoe and a vise. It’s sold for $569 on Amazon or $599 on the LUNYEE store, where you can also add the dust shoe ($45.99) and the CNC vise worktable ($69.99) among other options.

CNXSoft: This article is a translation – with some additional insights – of the original review on CNX Software Thailand by Suthinee Kerdkaew.

The post LUNYEE 3020 Nova review – A desktop CNC router with 710W spindle tested with UGS and Easel software appeared first on CNX Software - Embedded Systems News.