A new processing-in-memory (PIM) circuit developed by researchers at Washington University in St. Louis (WUSTL) adds neural network flexibility to PIM computing.
With PIM computing, the memory and processing unit are merged into a single unit, and computations are performed using the machine's physical properties.
The researchers are working on resistive random-access memory PIM, with resistors in both the memory and the processor.
Said WUSTL's Xuan "Silvia" Zhang, "In resistive memory, you do not have to translate to digital, or binary. You can remain in the analog domain."
To overcome the bottleneck that occurs when converting analog information into a digital format, the researchers added neural approximators that perform multiple calculations in the most efficient manner.
From Washington University in St. Louis
View Full Article
Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA