Real Processing in Phase Change Memory

Separation of processing and memory is the root cause of the main performance and energy bottlenecks in modern computers.

In the PI's ERC Stg, we proposed to combine data processing and storage in the same cells, to develop a novel unit called the 'memristive memory processing unit' (mMPU). This unit was based on resistive RAM (RRAM), which is still not commercial in large capacity.

A different memory technology, phase change memory (PCM), is commercially available in large capacity (for example, Intel's Optane has 1.5 TB memory), and can serve as an appropriate technology to form an mMPU with short time-to-market.In this PoC, we aim to experimentally demonstrate a working mMPU based on commercially-available PCM.This mMPU will enable fast and energy efficient computers that are cheaper than existing computers.

This will impact different applications such as artificial intelligence, databases, and genomics.

The mMPU-based computers will allow faster and cheaper execution of the aforementioned applications.In this proposal, we target the required technical and business steps required to develop such an mMPU and to commercialize it.