OUM is a rather nifty data storage technology, originally developed in the 1960s by Stanford Ovshinsky, aiming to provide a good compromise between the non-volatility of Flash, the speed of SRAM and the small size of DRAM.
Here's a quick rundown of the four technologies and how they work:
SRAM (Static Random Access Memory): Four transistors are coupled together to form a circuit that can be set to a stable value of 0 or 1 and then read back any number of times as long as the power is on. Switch it off, however, and you'll lose everything.
Another two transistors link the storage circuit to the buses. All address bits are accessed at once, and this gives very high performance but, with six trannies needed per bit stored, at a large cost of die space. Cache memory is typically made from SRAM.
DRAM (Dynamic Random Access Memory): Much smaller to make than SRAM, DRAM uses only one transistor in conjunction with a capacitor to store the value of the bit. This DRAM cell is common to most common memories: whether DDR2 SDRAM, DDR3 SGRAM, or even RAMBUS; it is the surrounding circuitry, the interfacing and mode of access that changes.
Because a capacitor is used to store the state, and these leak charge over time, the cell needs to be constantly refreshed. Most DRAM chips have a self-refresh mode that doesn't require the memory controller to do this; particularly useful for sleep modes, but remove the power again, and everything is lost.
However it is also slower than SRAM owing to the access mode being optimised for size over speed.
Flash: the most common form of NVRAM (Non Volatile RAM) these days, others include EEPROM and FRAM, Flash, of course, does not lose its memory when power is removed.
There are two types: NOR and NAND. Without going into too much detail, NOR Flash is generally used for smaller sized devices that need to be accessed via a straight address/data bus for executing code in place, such as a PC's BIOS, whereas NAND is accessed more like an IDE device, and is optimised for larger devices such as USB flash.
Both technologies rely on using a floating gate transistor to store the value. Effectively a capacitor is built into the MOSFET's gate, and another gate is added next to that to place or remove charge from it. Depending on how much charge is stored in this capacitor will determine what voltage is required to switch the transistor on or off.
This is stable and will remain valid over a long period of time, even without power.
Some Flash memories use a technology called MLC with a finer control over the amount of charge stored, and detection of the voltage needed to switch the transistor on, so can store more than one bit per cell.
Problems of flash include the limited number of times you can write to a cell (only about 10-100,000 cycles), and the fact that you can only write a zero value to a cell! The only way of restoring it to 1 is to erase the block of multiple cells fully, then re-write the whole lot. Of course this all happens without you knowing it!
OUM: The new kid on the block, pun intended, OUM uses a phase change alloy to store the bit's value, of a similar type to those used on recordable optical disks. Rather than having a transistor on or off to represent the 1 or 0, OUM toggles them between crystalline and amorphous states.
Crystalline material has a much lower resistance, being highly ordered, so it is fairly simple to tell what state it's in. And of course this doesn't change when the power goes, so it's non-volatile.
It's fairly easy to manufacture, as it just requires the deposition of a thin layer of metal onto the chip, resulting in fewer steps needed than flash memory, yet it's actually more reliable in the long term, lasting for more than 10 trillion write cycles. It is low voltage, fast, and is alleged to perform better when shrunk, unlike DRAM, where shrinking reduces the size of the capacitor unless you turn it vertical.
So it's a real all-rounder, but of course uses relatively untested materials and processors, so does not benefit from the years and billions of dollars poured into SRAM, DRAM and flash manufacture, so will be a while before you see it in commercial products.
Intel reportedly see this is a great technology for integrating onto the processor die itself, but for the time being have got their Flash memories group working hard on turning the Ovonyx concepts into a standalone product to compete with Flash.
We're pretty excited by the technology, as clearly are Intel, but also Elpida, Samsung, ST and Nanochip have all signed licensing agreements. Watch out for this in the iPods of the future. ยต