World's Speediest Flash Memory: Writes Data in Just 400 Picoseconds

Establishing a fresh standard for Flash memory efficiency, scientists at Fudan University in Shanghai have created an ultra-rapid, picosecond-level non-volatile memory component. Yet, what does picosecond-level memory entail? This term denotes memory capable of processing data reads and writes inside one-thousandth of a nanosecond—or essentially one trillionth of a second.

The recently created chip called "PoX" (Phase-change Oxide) can switch speeds as fast as 400 picoseconds, significantly outpacing the former global benchmark of 2 million operations per second.

Classic SRAM and DRAM can record information within a span of 1 to 10 nanoseconds. Nonetheless, these memory types are volatile; hence, once the power goes off, all saved data disappears.

In contrast, flash memory utilized in SSDs and USB devices maintains data integrity without needing constant power because it is non-volatile. However, this type of storage has significantly lower speeds, typically operating within the range of microseconds to milliseconds. Due to these limitations in read/write velocity, flash memory isn’t ideal for contemporary AI systems that require rapid movement and updating of extensive datasets throughout live operations.

As PoX is a form of non-volatile memory, it maintains data even when not powered down during periods of idleness. The blend of minimal energy usage along with extraordinarily rapid writing capabilities at the picosecond level might assist in overcoming the persistent memory constraint in AI equipment, wherein most energy expenditure currently goes towards transferring information instead of handling computations.

Professor Zhou Peng and his research group at Fudan University have entirely revamped the design of Flash memory. Rather than using conventional silicon, they implemented two-dimensional Dirac graphene, which is renowned for facilitating swift and unhindered movement of electrical charges.

They further refined the design by adjusting the Gaussian length of the memory channel, which allowed them to create a phenomenon known as 2D super-injection. This results in an exceptionally fast and nearly-unlimited flow of charge into the memory’s storage layer, effectively sidestepping the speed limitations that conventional memory faces.

"By using AI algorithms to optimize process testing conditions, we have significantly advanced this innovation and paved the way for its future applications," Zhou said in an interview with Xinhua .

To speed up the implementation of this technology in real-world settings, the research team is reportedly Collaborating closely with our manufacturing partners during the entire R&D phase. We have already conducted a tape-out verification which resulted in encouraging initial outcomes.

Liu Chunsen, who works at the State Key Laboratory of Integrated Circuits and Systems at Fudan University, stated, “We’ve successfully developed a compact, operational chip. Our upcoming phase entails incorporating this into current smartphones and personal computing devices. By doing so, using localized models directly on these gadgets should eliminate problems like sluggish performance and overheating which arise from limitations in present data storage methods.”

Follow Tom's Hardware on Google News To receive our latest news, insights, and evaluations directly in your feed, be sure to press the Follow button.

If you enjoyed this article, click the +Follow button at the top of the page to stay updated with similar stories from MSN.

Berbagi :