It may grab fewer headlines than processor design or RAM, but non-volatile memory has been a mainstay of Intel's product line for decades. As well as BIOS chips for PCs, the technology is ubiquitous in consumer electronics -- with the increase in digital media, world consumption has never been higher.
So far, the market is dominated by flash memories based on relatively standard silicon design processes, but a huge range of more exotic options is being explored.
Intel's flash memory technology, ETOX, was first developed 18 years ago but is still being aggressively improved. The transistors that store the memory bits have an insulated area: when a big enough voltage is put on one side of the insulation, electrons can leap across by a process of quantum tunnelling.
As they lack enough energy to tunnel out again when the voltage is removed, the area keeps this new charge. The electrons can be removed by raising a different area to a high voltage, whereupon they tunnel out again. Over the years, the voltages required have been reduced along with the size of the transistors, and the tendency of flash to wear out after too many writes has also been diminished.
Also, the write speed -- much slower than other memories -- has been gradually improved. Currently, ETOX is available at 0.13 micrometres -- like the rest of Intel's products, a move to 90 nanometres (nm) is planned, with a limit of 65nm envisaged.
System on a chip
The next major step after this point will probably be multi-bit memory cells: single blocks of memory that can store more than one bit of information. Multi-level cells can have one of four voltage levels stored in the same place, corresponding to two bits of information, while other techniques have two separate storage areas, one per bit. Intel also predicts the start of "System-On-A-Chip", where large amounts of flash memory are integrated onto the same die as a processor, interface circuitry and analogue components. This will have many advantages -- by eliminating external buses, it will use a lot less power and be less prone to noise, it will cost much less to produce and package, it'll be smaller and will work faster. In particular, Intel expects to place all the components of a wireless Internet device on one chip, combining XScale processor cores, Micro Signal Architecture radio communications and flash memory. Novel technologies
Beyond this point, Intel is looking at a wide range of novel technologies, including Polymer Memory, Ovionics Unified Memory (OUM), Magnetoresistive RAM (MRAM) and ferro-electric RAM (FeRAM). Each has its own characteristics which makes it more or less suitable for particular applications, but the company expects only one or two to make it into the mainstream. Polymer Memory is a zero-transistor-per-bit technology. It consists of layers of plastic polymer chains with a permanent electrical charge stacked between two sets of conductors, one running vertically and the other horizontal. By placing a voltage between one horizontal and one vertical line, the bit of the polymer between the two changes changes the polarisation of its permanent charge -- by sensing this polarisation through the same lines later, the memory bit can be read back. Polymer Memory isn't as fast as silicon memory but is much faster than disks. Multiple layers can be stacked on one chip to make a thick, capacious device; it needs little power and is very cheap to produce. It also has high capacity per dollar, and is seen as a good candidate for handheld data storage -- MP3 players, PDAs, digital cameras and the like. Its major drawback here is that reading the memory erases the contents, but refresh circuitry can solve this by immediately rewriting after reads. MRAM is much more complex. It uses a material that changes its magnetic polarity after a current is passed through it, and that changes its resistance in sympathy. This can be very fast -- almost as fast as DRAM -- but needs two transistors per bit. Nonetheless, it's got high density, doesn't wear out like flash and doesn't need refreshing after reads: the big challenge is making the magnetoresistive material compatible with standard CMOS design techniques. It fits neatly into the mainstream flash market, if costs can be brought down. FeRAM is a more exotic alternative. This relies on a compound called Perovskite, composed of crystals of a lead-zircon-titanium-oxygen molecule. A central atom can be moved between two positions by applying an electric field: this can be written and read very fast and needs little power; however, it has a limited life and like Polymer Memory suffers from a destructive read. Mass storage in the palm of your hand
Intel has high hopes for the last of its candidates, OUM. This is conceptually much simpler: a layer of chalcogenide alloy -- used in rewritable CDs and DVDs -- is placed above a mesh of electrodes. Power applied to one electrode heats it up, and this changes the alloy between its crystalline and amorphous states. Optical drives rely on the same property of the alloy but use lasers to heat up the material and effect the change, and to measure the change in reflectivity that results: in OUM, the change in the electrical resistance of the material is measured instead. The resulting memory cell is high density, doesn't have a destructive read cycle, needs little voltage and power and is easy to integrate with existing logic. It also has a lifetime of roughly a billion writes, making it a good candidate for mass storage in portable devices. OUM-like technologies have been studied for around 30 years: indeed, a paper co-authored by the industry legend Gordon Moore appeared in a technical journal in 1970, two years after he co-founded Intel. This gives the technology a certain cachet within the company, and 30 years of continual study and development of chalcogenide technology for optical storage have made it a much better understood material. Intel has demonstrated a 4-megabit test chip based on 0.18-micrometre design rules, and is producing a 0.13-micrometre version for further development. The company is very confident that one or more of the new technologies will be ready for production before ETOX begins to run out of steam. With demand for non-volatile memory expected to continue to grow as people use more and more digital data in their everyday lives, expect something exotic in your hand sometime in the next five years.