Keith Kressin, President and CEO
Edge AI plays a vital role in ensuring real-time data processing and ensuring high-performance data computation on IoT devices. However, most Edge AI solutions available in the market require tremendous configuration during implementation, and they are also not scalable.
MemryX, a fabless semiconductor company, aims to solve these challenges. Its Edge AI processing solutions powered by proprietary compute-at-memory technology and dataflow architecture set the bar in the industry. By combining high performance and accuracy with low system power, their solutions are fulfilling Edge AI needs for a number of industries, including transportation, IoT, and a wide range of industrial and consumer applications.
"Our innovative Edge AI solutions enable a single-click compilation and performance optimization of trained AI models, bringing superior performance, lower latency, and faster time compared to other solutions in the market." says Keith Kressin, President and CEO at MemryX.
MemryX was started with a vision to create a technology that can scale and quickly ramp customers with dedicated AI accelerators across a diverse embedded market. In 2019, at University of Michigan, the grand idea was materialized by Dr. Wei Lu, a professor of electrical engineering, with the support of his team.
The team jointly created a forward-looking core architecture from the ground up for robust software and hardware Edge AI solutions. With a focus on data-flow-centric AI that minimized the movement of data across memories, they created scalable solutions with customer-friendly features, that attach to any application processor along with the flexibility to configure future design requirements.
MemryX's team designed a highly efficient and scalable solution that is simple to deploy and support. The MemryX SDK allows developers to quickly develop systems and applications that optimally utilize MemryX AI accelerator chip(s). They are scalable to integrate with future-generation AI models, hardware, host processor, or operating system while eliminating all the bottlenecks of data movement. In addition, it supports a wide variety of different machine-learning models, or neural models, making it a highly programmable chip. The robust technology is available for high-performance AI devices, including automotive, robotics, drones, smart manufacturing, IoT, metaverse, edge computing, and smart vision system..
Our innovative Edge AI solutions enable a single-click compilation and performance optimization of trained AI models, bringing superior performance, lower latency, and faster time compared to other solutions in the market
MemryX's solution is beneficial for customers currently relying on CPUs or GPUs for AI processing. For example, one customer was dealing with poor speed and high latency of several AI models while running them on a CPU in a metaverse application. MemryX accelerated the data processing capability by migrating AI to the MemryX processor, reducing latency from 300 milliseconds to 30 milliseconds. In another instance, a customer was bearing the huge expense of large-scale power consumption of 250 Watts on a GPU for Edge AI computing. By migrating the AI models to MemryX, they achieved superior performance as well as lowering power consumption to just 10 Watts.
"In 2022 we have achieved broad customer sampling with positive feedback. After testing our solutions, customer have validated our assertions of ease-of-use with a powerful, yet power efficient solution that meets the needs of computer vision AI applications," says Kressin.
With company locations in the U.S., Taiwan, and India, MemryX is currently sampling the pilot silicon (MX3) and plans to enter production with an enhanced MX3 in 2023.