Microchip's analog embedded SuperFlash® technology helps storage and computing innovators

     Computing-in-memory technology is expected to eliminate the massive data communication bottleneck caused by artificial intelligence (AI) voice processing at the network edge, but requires an embedded memory solution that can simultaneously compute neural network and store weights . Microchip Technology Inc., through its subsidiary TPV Semiconductor (SST), announced that its SuperFlash memBrain neuromorphic memory solution addresses this challenge for WITINMEM neural processing SoCs. This is the first mass-produced SoC that enables sub-milliamp systems to reduce speech noise and recognize hundreds of command words in real time immediately after power-on.

      Microchip cooperated with Zhicun Technology to integrate Microchip's memBrain analog memory computing solution based on SuperFlash technology into Zhicun Technology's ultra-low power SoC. The SoC adopts integrated storage and computing technology for neural network processing, including speech recognition, voiceprint recognition, deep voice noise reduction, scene detection and health status monitoring. Zhicun Technology is currently working with multiple customers to bring products based on this SoC to the market in 2022.

      Shaodi Wang, CEO of Zhicun Technology, said: "Zhicun Technology is using Microchip's memBrain solution to solve real-time AI voice computing-intensive challenges at the network edge based on advanced neural network models. We pioneered the development of memory for audio in 2019. The all-in-one computing chip has now achieved another milestone, adopting this technology in batches in ultra-low-power neural processing SoCs, simplifying and improving the voice processing performance of smart voice and health products.

      Mark Reiten, Vice President of Licensing at SST Semiconductor (SST), said: "We are delighted that Zhicun Technology has become our key customer and thank Zhicun Technology for choosing to use our technology to launch such a good product and enter the expanding artificial intelligence edge processing market. The Zhicun SoC demonstrates the value of using memBrain technology to create a single-chip solution based on a memory-computing neural processor, eliminating the traditional processor's use of digital DSP and SRAM/DRAM-based methods to store and execute machine learning models. question."

      Microchip's memBrain neuromorphic memory products are optimized to perform vector-matrix multiplication (VMM) for neural networks. It enables processors for battery-operated and deeply embedded edge devices to deliver the highest possible per-watt AI inference performance. This is achieved by storing the weights of the neural model as numerical values in a memory array and using the memory array as a neural computing element, which consumes 10 to 20 times less power than other methods, while at the same time, since external DRAM and NOR are not required, the processor overall Bill of materials (BOM) costs are also lower.

      Permanently storing neural models in the processing elements of the memBrain solution also enables instant-on capabilities for real-time neural network processing. Zhicun Technology utilizes the non-volatility of the floating gate memory cells of SuperFlash technology to turn off the memory-computing integrated macro in the idle state, which can further reduce the static leakage power consumption of demanding IoT applications.