Princeton University researchers have constructed a programmable computer chip that boosts the performance and reduces power consumption of systems used for artificial intelligence (AI). The chip incorporates in-memory computing, eliminating the need for processors to retrieve data from stored memory, accelerating speed and efficiency.
The team combined in-memory circuitry into a programmable processor architecture, making the chip compatible with common coding languages like C. The chip supports systems designed for deep learning inference, and couples capacitors with conventional cells of static-random-access memory to perform computations on the data in the analog realm that are reliable and amenable to including programmability features.
Princeton's Naveen Verma says the device's energy savings are as important as its upgraded performance for AI applications that run on battery-driven devices. Programmability is necessary to ensure in-memory computing's capability will scale and be usable by system designers toward all desired AI applications, he says.
From Princeton University
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA