Artificial intelligence (AI) startup Mythic has raised a new round of funding to help develop its next generation M2000 processor.
The processor builds on the legacy of Mythic’s M1076 analog matrix processor and is 10 times more cost and power efficient than digital solutions. The M1076 processor is already shipping to customers such as Lockheed Martin and the company claims it is the first to achieve 33 ms latency on full HD high-accuracy object detectors.
The AI processors can be used for applications such as defense, enterprises, industrial environments, smart cities, smart home and more. Furthermore, it could also be adopted for computer vision applications like smart robots, security cameras, drones and augmented reality headsets.
The analog compute-in-memory (CIM) architecture of the AI processor uses analog memory to store neural networks completely on-chip and with density combined with integrated computation to deliver the same level of performance as a desktop GPU. This allows for edge AI inferencing in low-cost drones or tiny doorbell cameras but also robots and AR headsets.
The funding
The $13 million in funding was led by Atreides Management, DCVC and Lux Capital. Other investors include Catapult Ventures and Hermann Hauser Investment.
Mythic has raised $178 million in funding to date. It also announced that Dave Fick has been appointed to CEO after previously being CTO.
The company expects to use the funds to grow its headcount in 2024 and to help roll out the M2000 AI processor as production nears.
AI hardware rising
Mythic is not the only company that has seen growth in AI hardware. Nvidia is leading the charge with its AI processors being used for neural networks and machine learning (ML) applications as well as autonomous driving and advanced driver assistance systems (ADAS).
Syntiant Corp. recently expanded its line of special purpose AI silicon with neural decision processors designed for a wide variety of edge products. The company has been making deals with companies working on edge products including the recent always-on sensing development kit for ML/AI applications with Avnet.
While AI processing was once done mostly through software, the rise in hardware AI shows that in order to get the processing power needed for edge applications it will be done through dedicated processors.
