Page 28 - EngineerIT April 2022

ELECTRONICS

Computing-in-Memory innovator

solves speech processing challenges

SuperFlash memBrain™ memory solution enables WITINMEN’s System on Chip (SoC) to meet the
most demanding neural processing cost, power and performance requirements

omputing-in-memory technology is poised to eliminate memory neural processor that eliminates the problems of traditional
the massive data communications bottlenecks otherwise processors that use digital DSP and SRAM/DRAM-based approaches
Cassociated with performing artificial intelligence (AI) speech for storing and executing machine learning models.
processing at the network’s edge, but requires an embedded SST was founded in 1989, went public in 1995 and was acquired
memory solution that simultaneously performs neural network by Microchip in April 2010. SST is now a wholly owned subsidiary
computation and stores weights. Microchip Technology, of Microchip and is headquartered in San Jose,
via its Silicon Storage Technology (SST) subsidiary, memBrain California.
has announced that its SuperFlash memBrain As artificial intelligence (AI) Microchip’s memBrain neuromorphic
neuromorphic memory solution has solved processing moves from the cloud to memory product is optimised to perform
this problem for the WITINMEM neural the edge of the network, battery-powered vector matrix multiplication (VMM) for
and deeply embedded devices are challenged
processing SoC, the first in volume to perform Artificial Intelligence (AI) functions neural networks. It enables processors
production that enables sub-mA systems - like video and voice recognition. Deep Neural used in battery-powered and deeply-
to reduce speech noise and recognise Networks (DNNs) use AI applications that require embedded edge devices to deliver the
hundreds of command words, in real a vast number of multiply-accumulate (MAC) highest possible AI inference performance
time and immediately after power-up. operations to generate weight values. These per watt. This is accomplished by both
Microchip has worked with weights then need to be kept in local storage storing the neural model weights as values
WITINMEM, a leading provider of for further processing. This huge amount in the memory array and using the memory
computing-in-memory chips and system of data cannot fit into the on-board array as the neural compute element.
solutions, to incorporate Microchip’s memory of a stand-alone digital The result is 10 to 20 times lower power
memBrain analogue in-memory computing edge processor. consumption than alternative approaches, along
solution, based on SuperFlash technology, into with lower overall processor bill of materials (BOM)
WITINMEM’s ultra-low-power SoC. The SoC features costs because external DRAM and NOR are not required.
computing-in-memory technology for neural networks processing Permanently storing neural models inside the memBrain
including speech recognition, voice-print recognition, deep speech solution’s processing element also supports instant-on functionality
noise reduction, scene detection and health status monitoring. for real-time neural network processing. WITINMEM has leveraged
WITINMEM, in turn, is working with multiple customers to bring SuperFlash technology’s floating gate cells’ non-volatility, to power
products to market during 2022 based on this SoC. down its computing-in-memory macros during the idle state to
“WITINMEM is breaking new ground with Microchip’s memBrain further reduce leakage power in demanding IoT use cases. n
solution for addressing the compute-intensive requirements of
real-time AI speech at the network edge, based on advanced For information contact info@sst.com or visit the SST website.
neural network models,” said Shaodi Wang, CEO of WITINMEM.
“We were the first to develop a computing-in-memory chip for
audio in 2019, and now we have achieved another milestone with
volume production of this technology in our ultra-low-power neural
processing SoC that streamlines and improves speech processing
performance in intelligent voice and health products.”
“We are excited to have WITINMEM as our lead customer and
applaud the company for entering the expanding AI edge processing
market with a superior product using our technology,” said Mark
Reiten, vice president of the license division at SST.
“The WITINMEM SoC showcases the value of using memBrain
technology to create a single-chip solution based on a computing-in-

EngineerIT | April 2022 | 26