Microchip introduces SuperFlash memBrain
As artificial intelligence (AI) processing moves from the cloud to the edge of the network, battery-powered deep embedded devices are facing challenges in performing AI tasks such as computer vision and speech recognition. Microchip Technology Inc., through its subsidiary TPV, introduced the memBrainTM neuromorphic memory solution, a significantly lower power consumption, to effectively address this challenge. Microchip's analog memory solutions are based on industry-recognized SuperFlash® technology and are optimized for neural network vector matrix multiplication (VMM) execution, improving VMM system architecture implementation by simulating stored computational methods, and improving edge AI inference capabilities.
Since current neural network models may require 50M or more synapses (weights) to process, it becomes difficult to provide sufficient bandwidth for off-chip DRAMs, which becomes a bottleneck in neural network computing, and at the same time leads to an increase in overall computing power consumption. In contrast, the memBrain solution stores synaptic weights in an on-chip floating gate, significantly improving system latency. Compared to traditional digital DSP and SRAM / DRAM based methods, the power consumption of the new product is reduced by 10 to 20 times and the overall bill of materials (BOM) is significantly reduced.
Mark Reiten, vice president of SST Technology Licensing, said: "As technology vendors in the automotive, industrial and consumer markets continue to implement VMM for neural networks, our architecture can help increase the power, cost and latency of these proactive solutions. Performance. Microchip will continue to provide high reliability and versatile SuperFlash memory solutions for AI applications."
Companies looking to improve the machine learning capabilities of edge devices have begun adopting the memBrain solution. The memBrain analog memory computing solution is ideal for all AI applications due to its ability to significantly reduce power consumption.
Syntiant CEO Kurt Busch said: "Microchip's memBrain solution provides ultra-low-power storage computing for our upcoming analog neural network processors. We are in voice, image and other sensor modes on edge devices. Uninterrupted applications offer a variety of ubiquitous machine learning capabilities, and the partnership with Microchip brings many key benefits to SynTIant."
SST will showcase this analog memory solution at the 2019 US Flash Summit and showcase Microchip's flash performance extension architecture based on the memBrain product block array at the Artificial Intelligence/Machine Learning Forum. The 2019 American Flash Summit will be held from August 6th to 8th, 2019 at the Santa Clara Convention Center in Santa Clara, California.
For more information on the memBrain solution, please email [email protected]
SST provides design services for its memBrain solution and SuperFlash technology, and provides a software toolkit for neural network model analysis.