At school, I am sure that you learnt that computers and other digital equipment work on the basis of binary. This is almost exclusively true – except where flash memory is concerned. The very principle of MLC flash is that it allows the storage of two bits (ie numbers 0-3) in each cell as opposed to just one (binary – 0 and 1). How is this achieved? Let’s have another look at those buckets from earlier.

If we’re storing a binary value in a bucket, we’re essentially asking whether the bucket is full or not. If it’s full, it’s a 1, if empty it’s a 0. Over time, a cell might drift by gaining or losing electrons – essentially the equivalent of our bucket developing a slow leak or being rained on. Let’s say that the bucket after 3 weeks is 70% full. Is that a 0 or a 1? We can reasonably guess it’s a 1 as 70% is closer to 100% than to 0%.

With SLC, a cell is on or off. With MLC, it can be 0, 1, 2, 3.

With MLC, we’re storing four different levels in one bucket and therefore the margin for error is reduced – 70% full now equates to a value of 2 (or 10 in binary…). Therefore the ECC functions on the flash chip become much more important and the potential for corruption is much higher. It’s also much slower to see the value of a two bit gate – using our bucket analogy, it’s quicker to glance and see whether something is full or not, or empty, got a bit of water in it, quite full or full*.

MLC flash is a lot cheaper for the simple reason that it therefore has half the number of gates and therefore more chips can be made on a single die. As you can see from above, there is a significant tradeoff for this reduction in cost – reduced reliability and slower speeds .

Reductions in die size with CPUs and other processors has resulted in a significant increase in speed. With flash it’s had more or less the opposite effect. Using our bucket analogy, we’re talking about smaller and smaller buckets of water and therefore there are smaller and smaller differences in amount of water for each value. That means that for instance half an inch of rain might be an entire value difference as opposed to a tenth of one in a bigger bucket.

Now, we’re beginning to see TLC or Three Bit Flash hit the market – storing values 0-7 in each cell. Every single one of the problems above is maginified by a factor of two! That’s why we’re currently avoiding TLC products and will continue to do so until stability and reliability are considerably improved.

* a fairer analogy to how flash actually works would be to imagine the bucket sat on a lever that allowed varying amounts of water through a pipe depending on the weight of the water in the bucket…

Managing director, Dennis Ho-Young said “Although we’ve been selling these products into the IT market for nearly twenty years, with the convergence between the electronics and computing markets, we felt it was time to give them the focus and the specialist attention they deserve.

“We appreciate that whilst the approach used in the IT sector can help designers reduce costs, it often lacks the
specialist knowledge, advice and appreciation of industrial product lifecycles that the industrial market requires. This has meant that many companies developing new industrial and embedded products feel tied to the expensive and old fashioned electronics distributors. By aiming to combine the two, we hope to provide a service that will help our customers reduce costs whilst maintaining and improving the service they receive.”

The division has also launched a website at www.industrial-memory.co.uk where customers can find information on
a small proportion of the products and services offered. This will be continually enhanced over the next six months to provide a comprehensive resource.

A SIP - though not a memory chip - this is some kind of amplifier

Over time, memory has come in various shapes and sizes. Suffice to say the main reason that you need to buy the right format is that it won’t fit if you don’t!

A DIP chip

Back in the mists of time, memory came as loose chips, normally in what are called DIPs or SIPs  – Dual in Line Package or Single in Line Package – which were plugged either directly on to the motherboard or into a large board plugged into an ISA slot or the like. On some machines (for example the Atari ST) the DIPs had to be soldered directly to the board which was a laborious and error prone process.

Later, manufacturers decided that the ability to easily add and remove memory would be beneficial for both them and their users – it would allow the manufacturers to assemble the machines more quickly and offer more configurations, and would permit users to easily add more memory later.

A 30 pin SIMM

Therefore, after a bit of faffing about with modules such as SIPPs (not to be confused with SIPs),  a simple standard for a “module” that would support a common interface was adopted – the 30 pin SIMM – the Single Inline Memory Module.

A 72 pin SIMM

As time went on the memory busses on computers became wider and more than one SIMM was required to support the amount of memory the computer could see. By the time of the 486, four SIMMs at a time would have to be installed, and with the advent of x8 chips, it was becoming silly to have so many SIMMs on a motherboard. Therefore the 72pin SIMM was invented. This, in effect, is four 30-pin SIMMs rolled into one, and supports a width of 32 or 36 bits rather than 8 or 9 (see previous post on parity and ECC).

Around about the same time that the industry moved to SDRAM from EDO, a similar move was made to move to 64 bit widths, mainly due to the larger numbers of ancilliary (non-data) connections required to talk to SDRAM. This required a move to a Dual Inline design as fitting 168 pins along the bottom of the module at 5V would have required a memory module nearly seven inches across!

Over time, more pins have been added as the voltage has dropped and more separation has been required between data signals, but essentially the format has remained similar.

What’s the difference between a SIMM and a DIMM?

A DIMM

A single inline module, although having pins on both sides of the module, effectively only has one set of pins as the two sides are connected together, whereas a DIMM has totally separate sets of pins on the two sides and hence can fit twice as many pins in the same space.

So what’s a SODIMM then?

DDR2 SODIMM

A SODIMM or Small Outline DIMM is just a smaller form factor of memory made to fit in confined spaces. In order to save on space, it doesn’t have the pins to support functions like ECC and parity and tends to have smaller gaps between the pins.

Over time, SODIMMs have been available in 144pin (SDRAM), 200pin (DDR/DDR2) and 204pin (DDR3) formats, though other, specialist types like 100pin (Printer DIMMs) and 172pin “MicroDIMMs” (even smaller) also exist.

And a RIMM?

A RIMM is just a trade name for a specific type of DIMM which carries RAMBUS memory. There’s nothing special about them except that they are totally incompatible with any other kind of memory.

So what are the notches for?

Most DIMMs have a notch or notches in the bottom between the pins. These serve three purposes:

1. To prevent the wrong memory being plugged into a slot where it could damage the computer or the memory.

2. To identify the memory technology. On 168 and 240pin DIMMs and 200pin SODIMMs, the notch identifies the type of memory – EDO or SDRAM for 168pin, and DDR3, DDR2 or DDR for the 240pin and 240pin parts.

3. To identify the voltage on 168pin DIMMs and 144pin SODIMMs as this changed between 5V and 3.3V without changing technologies. On newer parts, voltage changes were made at the same time as the technology changed.

How Parity Works

By the era of 30pin (8 bit) SIMMs, it was essential to be able to detect if and when this happened. Therefore, rather than use 8 x 1bit wide chips on a module, the designers added a 9th and stored a parity bit for each byte in it. This allowed single memory errors to be easily detected as the parity bit would be wrong. This practice continued with 72 pin (32bit) SIMMs which had a parity bit added to each byte (ie a parity SIMM was 36 bits wide).

High end manufacturers like Sun and SGI had by this time found the two primary limitations of Parity on memory modules:

1. Parity can only detect a single error – if there are two or more errors, the Parity bit stands a 50% chance of being correct, and the error will go undetected. This wasn’t a major problem until memory ICs more than one bit wide started appearing, but rapidly became one.
2. Parity can only detect errors – it can’t correct them.

Therefore these manufacturers put a second parity chip on custom modules and used it to detect parity on each 16bit combination of two adjoining bytes. This solved the first issue but not the second.

ECC modules have an odd number of chips - in this case 9.

By the advent of DIMMs, more computing power was available and there was the potential for a new system. Rather than using the extra one bit per byte for storing a simple parity bit, the extra byte was instead used to store a Hamming code for the entire 64 bit word. This allowed single bit errors to be corrected and dual bit errors to be correctly detected.

Unless something goes physically faulty, it is therefore unlikely that the system will encounter an error that will cause it to crash – and indeed the only time that most users of machines with ECC will know that it’s encountered an error is if they check the BIOS log.

There still remain, however, a major of issue that ECC memory doesn’t resolve. Should more than one bit in a byte disappear, it’s still reset time – and individual chips are more likely to see two bit errors than two bits acquiring errors in separate chips.

Interleaved memory distributes the bits of each byte to multiple chips and or modules to increase resilience.

Interleaving is one way of reducing the chances of this happening. This is where individual bytes are written to one or more modules in a different order to what one would expect in order to reduce the chances of irrecoverable errors.

Advanced ECC (or Chipkill in HP parlance) is an extension of this principle which relies on using “by 8″ chips (so the module is exactly 9 chips wide) and inteleaving both the data and the Hamming codes. This allows the failure of an entire chip on the module without the loss of any data.

Ultimately, for most users, ECC is of limited benefit as the chances of encountering recoverable memory errors is minutely small. However, when data is critical, such as in a server or high end workstation, the extra cost is easily justified for the security it brings.

Printer memory comes in all shapes and sizes

Whilst almost everyone knows the benefits of upgrading the memory in PCs, laptops and servers, not so many are aware just how much difference upgrading printer memory can make.

If you’re seeing printer “Out of Memory” errors or printer “Page too big” errors, that means you need to upgrade your printer memory. Also, if your printer pauses sometimes for no apparent reason or if it doesn’t go as fast as it ought to, a printer memory expansion could be the answer.

A more typical printer DIMM

But Printer Memory is expensive – manufacturers like HP and Lexmark charge up to £550 for a module. Here at Memory Express, we’ve been selling a wide range of printer memory upgrades for nearly twenty years and our current range is second to none. Best of all, our memory is up to 80% cheaper than the OEM so you can save a fortune! You’ll find most models listed on our Memory Configurator; if you can’t find yours, give us a call.

To find Printer Memory Upgrades click here.

A fully buffered dimm

One of the perennial challenges in memory for the last decade has been fitting enough slots on a motherboard to ensure users can fit enough memory on to fulfill their needs. At first glance, this seems like a silly problem – surely more sockets can just be soldered on?

The problem lies in that a conventional setup requires a separate set of connections is required to connect the south bridge (or processor depending on your setup) to each memory module. There are about 160 data, control and address lines on a 240 pin DIMM, so for four sockets, we’re talking about 640 leads to be routed through the circuit board to the right place in a very small amount of space. Just to make things even more complicated, all the leads must be exactly the same length or some of the signals will arrive before others; this already results in loads of wiggly lines on the circuit board as the designer strains to keep the distances equal.

When DDR2 arrived on the market, the biggest module size available was 2GB but many servers needed 16GB or even 32GB of memory. Now, with our one-set-of-leads-to-each-module scenario and either 8 or 16 slots, we’re talking about crazy numbers of connections: 1280 or 2560 – and that’s without the power or earthing!

With this in mind, the industry started looking for solutions to this problem. JEDEC arrived at a surprisingly simple one- using similar technology to SATA disks, a controller chip is placed on each DIMM and then the data is sent over a smaller number of serial buses back to the memory controller. The other advantage of this is that each memory bank then only requires enough leads for differential driven data and control lines – meaning that as few as sixty connections are needed per channel.

This serial line controller is known as a Serial Line Buffer. To avoid confusion with ordinary registered (buffered) modules, these are hence called Fully-Buffered DIMMs.

Banks

Two banks - two colours

A memory bank is normally either one or a set of sockets that must be populated together. That’s why, on our configurator, we list both how many banks a machine has and the number of modules in a bank.

Normally in desktop machines, you’ll see one (low end machines), two (dual channel machines) or three (Core i7 machines) modules in a bank, though some servers with multiple processors may take modules four, six or eight at a time; also, when SIMMs were still current, you often saw banks of four in desktops.

In most machines, each bank has its slots coloured differently – if you see two white and two red slots on the motherboard, you’ll need to install memory in both the slots of one colour.

Ranks

A quad rank module consists of four "virtual" modules.

I don’t know which clown decided to call ranks “ranks” but he (or she) has a lot to answer for. I’ve always preferred to think of them as virtual modules instead. Nonetheless, this concept haunts machines running FB-DIMMs and DDR3 server DIMMs so we’d better have a look in to it…

The big problem comes when a chipset supports a number of ranks that is less than could be installed using common modules. For example, there are a few Intel server chipsets that support, for instance, eight slots but only 12 ranks supported. This means that installing eight dual-rank modules will not work as all the ranks have been used up by the first six slots.

You’d think, therefore, that using modules with the lowest number of ranks would always be the best idea. However, just to make things a bit more brain-addling, there are machines that won’t support dual rank 4GB modules and will only read quad rank ones.

Overall, the lesson from this is not to worry about it and to trust the configurator! Twenty years of knowledge and the correct memory choices for 40,000 machines mean that we can look after all the difficult bits. And we will always call you if you do manage to order an uninstallable combination!

Sides

8 chips of 128M x 8 = 1GB

Once upon a time, memory modules would either have chips soldered on both sides or just on the front – these were known as double or single sided modules. Fast forward fifteen years and some “double sided” modules only have chips on one side.

To get to the bottom of what people mean by n-sided modules, you need to think about how the memory is arranged within the module. Sometimes you’ll see a normal 1GB desktop DIMM described as a 128M x 64 module. What this means is that the memory is 64 bits wide (almost all DIMMs are) and 128 million (or more accurately 2^27) bits deep.

Equally two sides each of 8 x 8x64M chips makes 1GB

If you consider this as a piece of land, there are lots of ways this can be divided into equal sized fields (or in reality memory chips):

• 4 chips of 128M x 16
• 8 chips of 128M x 8
• 16 chips of 128M x 4

or… we could put two rows of fields in:

• 2 rows of 8 chips of 64M x 8
• 2 rows of 16 chips of 64M x 4

Those two rows of chips are the “sides” of which we speak. In essence, they’re pretty similar to the ranks mentioned above, but the whys and wherefores of whether they’ll work in a given machine is rather different. Once again, we’d strongly recommend you simply don’t worry about it and trust the configurator!

• request quotations
• convert quotes to orders
• track shipments from Memory Express

We strongly recommend customers to run Microsoft’s newer Windows 7 operating system as it is more secure, faster and will be supported for many years to come.

Machines eligible for downgrade

Computers supplied with a Windows 7 Professional or Ultimate license (ie an “OEM” copy) can be legally downgraded to any previous Windows Professional OS (ie NT4, Win2K, XP Pro, Vista Business).

Where can I get an XP disc?

You will need a generic Windows XP disc from another machine to install. We can no longer supply XP discs.

Installing XP Pro and obtaining a key

Windows activation screen

Boot from your XP disc and install as normal; if you get a “Blue Screen of Death” on first boot, don’t panic and please see the section on installing on SATA machines below.

When asked for a license key, enter one from another of your machines (don’t worry – we’ll change it shortly) and proceed with the install

When you reach the activation screen, select “Activate by telephone” and dial the 0800 number on the screen. Press the option given to speak to a customer service rep. When they answer, tell them you’re downgrading from Windows 7 to XP Pro and read them the Windows 7 key. They will then give you another code to type in to the machine and you will be up and running.

Installing Windows XP on SATA Laptops and Desktops

A typical BIOS AHCI mode option

With many Windows XP discs now being over 5 years old, it is unlikely that it will contain the drivers for your SATA controller, without which it is not possible to install. Although it is possible to load the drivers using F6 on boot, PC makers are aware of this issue and have supplied an easy workaround.

Before installation, enter the BIOS setup screen, normally by pressing DEL on boot (some machines may require F2, INS or F12 to be pressed instead). Look for an option called “Configure SATA as” or “SATA mode” or “AHCI mode”. Change this setting to “IDE” or “ATA emulation”. This will cause the SATA controller to emulate an older IDE interface making the drives available to XP.

When you have finished installing the machine, install the SATA drivers and return to the BIOS. Change the setting back to “AHCI” and reboot. You will have access to your drives at full speed.