Investors Bet Flex Logic Embedded FPGAs will Catch Edge AI Wave With $55M New Funding
Geoff Tate, CEO and cofounder of Flex Logix Technologies, told us several years back that edge AI inferencing would be the embedded FPGA vendor’s biggest market through about 2023.
Among the reasons was the crying need for an affordable inference engine that could also handle a range of applications as data processing and storage shifts from cloud data centers to edge devices.
Additional evidence supporting that prediction emerged this week: The intellectual property specialist closed an oversubscribed $55 million funding round led by Mithril Capital Management. Existing investors Eclipse Ventures, Lux Capital and the Tate Family Trust also chipped in to the Series D round.
Earlier funding rounds raised a combined total of $27 million. The company’s current valuation was not disclosed.
Flex Logic said Monday (March 22) it would use the funding to expand its applications for AI inference and new embedded FPGA uses cases as the enterprise edge market seeks affordable yet high-performance inference engines.
“Embedded FPGA is in demand by customers to enable their chips to adapt to new algorithms and protocols,” said Tate, who previously served as founding CEO of chip IP pioneer Rambus Inc.
The company’s embedded FPGA architecture seeks to provide faster programmable hardware interconnects requiring less memory bandwidth. Among other advantages, that framework reduces DRAM bandwidth requirements—fewer DRAMS translates to lower cost and less power for edge applications.
Flex Logic’s InferX X1 platform is optimized for edge applications such as megapixel machine vision use cases requiring low latency. The architecture combines Tensor processors with a reconfigurable interconnect that boosts resource utilization within each layer of a neural network. Those high-speed connections are used to reconfigure computing and memory as models are processed.
The company’s IP cores support the latest version of its edge inference accelerator running version 3 of the YOLO (as in, “you only look once”) real-time object detection algorithm. The InferX X1 accelerator is currently sampling with early customers. The company said chips and boards will be available by mid-year.
Meanwhile, Flex Logic is moving along a parallel software track, making its inference compiler generally available this summer. The compiler accelerates high-level neural network models, then generates the code to run its InferX X1 “without the detailed programming other solutions require,” Tate said.
With venture capital in hand, he said Flex Logic’s embedded FPGA technology addresses an edge AI market ripe for expansion as more applications emerge and companies look for cheaper alternatives to pricey GPUs. Eschewing the crowded data center market, Tate said “the edge enterprise market is a place that has a need” for affordable, high-end inference engines.
Tate also noted in an interview that it took other IP core vendors like Arm years to gain market traction. “It’s still early days, but once you cross that threshold,” the edge AI market will help drive demand for embedded FPGA hardware and software, he predicted.
Positive cash flow from the company’s embedded FPGA business has allowed Flex Logic to operate until now with just $27 million in venture funding. Investors are betting the edge AI market is poised to explode while SoC designers integrate FPGAs into data center ICs.
Lead investor Mithril was founded by Ajay Royan and Peter Thiel, the co-founder of Palantir Technologies. In announcing the investment, Royan noted the embedded FPGA vendor’s push into edge enterprise inference markets ranging from medical and retail to industrial and robotics.
Among edge applications, Tate called robotics an “exponential-growth market.”
George Leopold has written about science and technology for more than 30 years, focusing on electronics and aerospace technology. He previously served as executive editor of Electronic Engineering Times. Leopold is the author of "Calculated Risk: The Supersonic Life and Times of Gus Grissom" (Purdue University Press, 2016).