Syntiant, a start-up that develops AI edge hardware for voice and sensor solutions, today completed a $35 million financing. Chief executive Kurt Busch said the money would be used to ramp up production for the rest of 2020.
The market for voice and speech recognition hardware is expected to reach $26.8 billion by 2025, according to a report by ConocoSe Research. This is because devices such as smart speakers, smart displays, mobile phones, headphones, hearing aids, and laptops require low-power chips to handle statements. While some system-level chip products use coprocessors to handle speech recognition, they often do not fit into multiple form factors.
Headquartered in Irvine, California, Syntiant, a three-year-old company, provides hardware that combines machine learning with semiconductor design for always-on voice applications. The company’s neural decision processor with Alexa voice service firmware provides wake-up words, command words, and event detection in a package known as “near zero” power consumption. In fact, Syntiant says its processor throughput is 20 times that of a competitor’s microcontroller, making the device about 200 times more efficient.
Syntiant’s NPD100 and NPD101 processors are approximately 1.4 mm x 1.8 mm in size and can run models with more than 500,000 parameters. The two chips package a universal ARM Cortex-M0 processor with 128KB of memory, consume less than 140 microwatts, and enable on-board firmware security and certification, keyword training, and up to 64 output classifications.
Syntiant’s N100 and N101 initially aimed at the performance of about 20 TOPS (trillion floating-point operations) per watt, using hundreds of thousands of flash NOR units to read and write one word or a byte of data at a time. The in-memory processor architecture was proposed by Jeremy Holleman, a researcher at the University of North Carolina in Charlotte, at the 2014 International Solid State Circuits Conference. Syntiant claims that the architecture is ideal for large-scale parallel operations in deep learning with low power consumption.
Syntiant unveiled the prototype for the first time at Infineon’s OktoberTech conference in 2018. The company began shipping customer samples later that year and shipping the first products in September 2019, one month after receiving Amazon’s Alexa hardware certification. Syntiant’s neural decision processor comes with a software development kit (SDK) that integrates with virtually any integrated development environment, as well as a “Training Development Kit” (TDK) that allows developers to use Google’s TensorFlow machine learning framework and other set of machine learning tools.
M12 (formerly Microsoft Ventures) and Applied Ventures, a company that invests in this week’s round of C financing, with new investors Atlantic Bridge Capital, Alpha Edison and Miramar Digital Ventures also taking part. That brings the company’s total financing to more than $60 million. As part of this round of financing, Michael Stewart, investment director at Applied Ventures, joined Syntiant’s board of directors as an observer of the board along with David Lam of Atlantic Bridge Capital.