AI processing moves from the cloud to the edge.
Optimized Neural Network VMM Based on SuperFlash Technology
Membrane Solutions Improve System Latency

As AI processing moves from the cloud to the network edge, high-performance, battery-powered embedded devices are faced with the challenge of performing AI functions like computer vision and speech recognition.

Microchip Technology, through its subsidiary Silicon Storage Technology, announced on the 12th that it can address this situation by applying its analog memory technology, the memBrain neuromorphic memory solution, to significantly reduce power consumption.

Microchip's analog flash memory solution, optimized for performing vector-by-matrix multiplication (VMM) in neural networks based on SuperFlash technology, enhances the system architecture implementation of VMM through an analog in-memory computing approach, elevating AI inference capabilities on edge devices.

Current neural network models require more than 50 million synapses (weights) for processing, making it difficult to secure sufficient off-chip DRAM bandwidth. This has created a bottleneck in neural network computing, increasing overall computing power consumption.

In contrast, membrane solutions are synaptic weightedSystem latency is improved by storing the chip in on-chip floating gates. Compared to conventional digital DSP and SRAM/DRAM-based approaches, the membrane solution reduces power consumption by 10 to 20 times and the bill of materials (BOM).

“As technology manufacturers targeting the automotive, industrial and consumer markets continue to implement VMMs for neural networks, SST’s architecture helps these forward-thinking solutions achieve power, cost and latency benefits,” said Mark Reiten, vice president of SST’s licensing business unit. “Microchip continues to deliver reliable and versatile SuperFlash memory solutions for AI applications.”

“Microchip’s memBrain solution will enable ultra-low-power in-memory computation for our upcoming analog neural network processors,” said Kurt Busch, CEO of Syntiant. “Our partnership with Microchip continues to provide Syntiant with several important benefits as we support holistic machine learning for always-on applications in voice, image and other sensor modalities on edge devices.”