Flex Logix raises $55M to design AI chips for edge enterprise functions


Be part of Rework 2021 for a very powerful themes in enterprise AI & Knowledge. Learn more.


Flex Logix, a startup designing reconfigurable AI accelerator chips, right now introduced that it closed a $55 million funding spherical led by Mithril Capital Administration. CEO Geoff Tate says the funding will allow the corporate to construct out its software program, engineering, and buyer help groups to speed up the supply of its {hardware} and software program for edge enterprise functions.

AI accelerators are a sort of specialised {hardware} designed to hurry up AI functions, significantly neural networks, deep studying, and numerous type of machine studying. They give attention to low-precision arithmetic or in-memory computing, which may enhance the efficiency of huge AI algorithms and result in state-of-the-art ends in pure language processing, pc imaginative and prescient, and different domains. That’s maybe why they’re forecast to have a rising share of edge computing processing energy, making up a projected 70% of it by 2025, according to a latest survey by Statista.

Mountain View, California-based Flex Logix, which was based in 2014, claims its AI inference chip — InferX X1 — is among the many quickest and most effective. The InferX1 outperforms Nvidia’s Xavier NX on the favored pc imaginative and prescient benchmark YOLOv3 and “actual buyer fashions,” in response to Flex Logix, and the corporate says it’s focusing on a price-to-performance ratio 10 to 100 instances higher than present edge inference options.

“Flex Logix got down to be for FPGA what Arm is for processors,” Tate instructed VentureBeat by way of e-mail. “We imagine this authentic eFPGA enterprise can develop to be as large as Arm’s over time, whereas our second line of enterprise is driving edge AI Inference capabilities into excessive quantity functions, thus rising the market to the billions of {dollars} that market forecasters predict.”

The InferX X1 additionally options what Flex Logix calls a reconfigurable tensor processor, nnMax, containing 64 processors coupled with SRAM that may be reprogrammed in 4 millionths of a second. In machine studying, a tensor is a generalization of vectors and matrices — representations of the info inputs, outputs, and transformations inside neural networks. Flex Logix asserts that the nnMax is 3 to 18 instances extra environment friendly by way of throughput per millimeter squared than the typical Nvidia GPU.

“[The nnMax] reconfigures the 64 [processors] and RAM assets to effectively implement a layer with a full bandwidth, devoted knowledge path, like an ASIC, then repeats this layer by layer,” Flex Logix explains on its web site. “[We use] a brand new breakthrough interconnect structure with lower than half the silicon space of conventional mesh interconnect, fewer metallic layers, greater utilization, and better efficiency … We are able to simply scale up our architectures to ship compute capability of any dimension … utilizing a patented tiling structure with interconnects on the fringe of the tiles that routinely kind a bigger array of any dimension.”

On the software program facet, Flex Logix’s compiler takes fashions from machine studying frameworks together with Google’s TensorFlow and ONNX and optimizes them for its nnMax and InferX1 architectures. A efficiency modeler is obtainable now and in use by “dozens” of shoppers, and Flex Logix finally plans to make accessible software program drivers for working methods generally utilized in server and real-time situations.

Flex Logix’s merchandise have but to come back to market, however after they do, the corporate says they’ll be accessible in PCIe card and M.2 format for edge servers and gateways. A PCIe board containing the InferX1, X1P1, is anticipated to kick off manufacturing in Q2 2021 priced between $399 and $499, relying on the processor velocity. A much less highly effective variant of the chip, InferX1 1KU, will value between $99 and $199, with quantity pricing reaching as little as $34 to $69.

Flex Logix

Flex Logix has competitors in a market that’s anticipated to succeed in $91.18 billion by 2025. In March 2020, Hailo, a startup creating {hardware} designed to hurry up AI inferencing on the edge, nabbed $60 million in enterprise capital. California-based Mythic has raised $85.2 million to develop customized in-memory compute structure. Graphcore, a Bristol, U.Ok.-based startup creating chips and methods to speed up AI workloads, has a conflict chest within the a whole bunch of thousands and thousands of {dollars}. And Baidu’s rising AI chip unit was recently valued at $2 billion after funding.

However Flex Logix investor Ajay Royan factors to Tate’s pedigree as one motive for his continued confidence. Tate beforehand managed AMD’s microprocessor and logic group, and he took his first startup, chip licensing agency Rambus, from 4 folks and $2 million in fairness to a Nasdaq IPO and multibillion-dollar market cap. Flex Logix says its income in 2020 was within the “double digits” thousands and thousands and is anticipated to develop 50% to 100% this yr.

“We’re impressed with the … structure that Flex Logix has developed based mostly on distinctive mental property that offers it a sustainable aggressive benefit in a really excessive development market,” Royan mentioned in a press launch. “This expertise benefit positions Flex Logix for fast development in edge enterprise inference in functions corresponding to medical, retail, industrial, robotics and extra. It’s much more spectacular that they’ve performed this with so little capital and on the similar time constructed a cash-flow constructive … enterprise with giant development potential as system-on-chip designers look to include reconfigurability into their communications and knowledge facilities.”

Lux Capital, Eclipse Ventures, and the Tate Household Belief additionally participated in Flex Logix’s newest fundraising spherical, a sequence D. It brings the corporate’s complete raised up to now to $82 million.

VentureBeat

VentureBeat’s mission is to be a digital city sq. for technical decision-makers to realize data about transformative expertise and transact.

Our web site delivers important data on knowledge applied sciences and methods to information you as you lead your organizations. We invite you to change into a member of our group, to entry:

  • up-to-date data on the themes of curiosity to you
  • our newsletters
  • gated thought-leader content material and discounted entry to our prized occasions, corresponding to Transform 2021: Learn More
  • networking options, and extra

Become a member



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *