New chip boosts AI computing efficiency
The battery power is what limits the AI capabilities of these small edge devices. It is therefore crucial to improve energy efficiency. Data processing and storage are done at different places in today’s AI chip – the compute unit and memory unit. Data movement between the two units is what consumes the most energy when AI processing takes place.
Stanford University engineers are working on a solution. They have developed a new resistive random-access (RRAM), which integrates AI processing into the memory, eliminating the need for separate compute and memory units. The \”compute in memory\” (CIM), called NeuRRAM is the size of a small fingertip. It can perform more tasks with less battery power.
H.-S Philip Wong is the Willard R. & Inez Kerr bell Professor at the School of Engineering. He said that by storing the calculations on the chip, rather than sending the information to the cloud, it could make AI faster, cheaper, safer, and more scalable in the future.
Source:
https://news.stanford.edu/2022/08/18/new-chip-ramps-ai-computing-efficiency/
Leave a Reply