Mythic’s platform delivers the power of desktop GPU in a single chip, low power form factor, supporting large deep neural networks for inference in high-value embedded applications. It offers massive parallel computing when needed, then vanishes when not. The result is near-weightlessness of device constraints – all the inference capacity needed to support complex, large deep neural networks, with little of the overhead and complexity of traditional solutions. Mythic’s technology is based upon an entirely new hybrid digital/analog flash calculation using 8-bit non-volatile memory arrays which has been under development since 2012. The architecture supports all popular styles of deep neural networks, runs deterministically, and can be fully optimized for a broad variety of applications. Because weights are stored on-chip, Mythic’s hardware interface and driver model are simple – point to the data, trigger inference, and read the results. For especially demanding applications, multiple Mythic chips can chained together and parallelized to provide compute at a large scale impossible with existing solutions.
Mythic
Recent Content by Company
“Pioneering Analog Compute for Edge AI to Overcome the End of Digital Scaling,” a Presentation from Mythic
Mike Henry, CEO and Founder of Mythic, presents the “Pioneering Analog Compute for Edge AI to Overcome the End of Digital Scaling” tutorial at the May 2019 Embedded Vision Summit. AI inference at the edge will continue to create insatiable demand for compute performance in power- and cost-constrained form factors. Taking into account past trends, …
“Mythic’s Analog Deep Learning Accelerator Chip: High Performance Inference,” a Presentation from Mythic
Frederick Soo, Head of Product Development at Mythic, presents the “Mythic’s Analog Deep Learning Accelerator Chip: High Performance Inference” tutorial at the May 2018 Embedded Vision Summit. This presentation explains how Mythic’s deep learning accelerator chip uses a unique analog circuit approach to deliver massive power, speed and scalability advantages over current generation deep learning …
