Pox, new storage technology 10,000 times faster
A team of researchers from Fudan University in China presented a new type of experimental RAM (RAM). Called Pox, this type of storage displays performance theory far superior to current technologies. This memory would indeed be characterized by a particularly high writing speed, capable of reaching 400 picoseconds per operation. This would correspond to a completely crazy theoretical flow of around 25 billion operations per second. The designers estimate that it would be potentially 10,000 times faster than the flash memory used in most laptops today.
The problem with current memories
The traditional traditional memories such as the DRAM and the SRAM already offer fairly substantial flows. Their performance has increased in concert with processors’ requirements, also more and more bicycle. However, the great weak point of these memories is their volatility. The data is lost as soon as the power supply is cut, which requires fully restarting the systems and recharging applications at each switch. For its part, the flash memory, which is found in SSD discs and USB keys, has the advantage of being non-voltile. She is able to keep the information recorded even off. But she also suffers from an Achilles heel: her reading speeds and especially writing are significantly lower than those of the Dram or the Sram. This (relative) slowness constitutes a bottle of strangulation for uses handling large volumes of data, such as artificial intelligence (AI) or the analysis of massive data (Big Data). These areas indeed require very fast memory access. This is where POX technology, which would be able to combine the speed of RAM and the non-voltility of flash memory, between on track.
Grave for speed and non-voltility
To achieve such a result, the specificity of PoX memory developed in Fudan is due to the use of two-dimensional graphene, a material with specific electronic properties, in place of silicon usually used in semiconductors. Grave is a form of carbon made up of a single layer of atoms arranged in a hexagonal network.
According to Fudan researchers, it is the effectiveness of ballistic load transport in graphene that makes a phenomenon baptized “super-injection” possible. A mechanism that would inject electrical loads into the memory cell in a much faster way than with current solutions. It is therefore thanks to this speed of load transfer that PoX memory would be able to offer ultra -curted writing times of 400 picoseconds.
The research team, led by Professor Zhou Peng, also said that he had recourse to artificial intelligence during the design phase to refine the design of PoX memory and optimize their theoretical performance. The results of this work have been published in the scientific journal Nature.
Potential advantages
In addition to its announced speed, the other asset highlighted for PoX technology is of course its non-voltile nature. Unlike DRAM and SRAM memories, the POX is designed to keep the data stored even when the device is off, without requiring continuous power supply. This characteristic could have an impact on the energy consumption of electronic devices.
The data centers hosting artificial intelligence servers, for example, consume large quantities of electricity, partly due to the need to constantly maintain large quantities of RAM. A non-voltile and rapid memory like the pox could potentially reduce this consumption. This would allow mobile devices such as smartphones and laptops to benefit from prolonged autonomy.
This non-voltility would also make possible a category of instant start-up devices (“Instant-on”). By keeping their condition, even extinct, these devices could resume their operation immediately after their switching on. In addition, fleas dedicated to artificial intelligence could be designed without the voluminous and energy -consuming sram cache currently necessary to compensate for the relative slowness of main storage memories. The integration of a quick and non-voltile memory like the pox directly on the chip could simplify their architecture and improve their efficiency.
Questions remain
If the announced performance of Pox memory are very interesting on paper, several points remain to be clarified before considering a large -scale adoption. To start, the Fudan University team has not yet communicated data concerning the holding of this technology over time. However, you can imagine, aspects such as endurance, that is to say the number of writing and erasure cycles that memory can support, or reliability on duration are obviously crucial points. This technology will also have to prove its mass production potential, at a reasonable cost. Because the transition from a prototype to industrial production is a pitfall against which laboratory researchers have encountered.
