Syntiant recently submitted its NDP120 into the MLCommons’ MLPerf Tiny category for keyword spotting, which is intended for the lowest power devices and smallest form factors, such as deeply embedded, intelligent sensing and IoT applications.
“Syntiant’s architecture yielded the industry’s lowest-power solution for running edge-based AI applications,” said Jeremy Holleman, chief scientist at Syntiant responsible for the company’s MLPerf submission. “Results of the MLPerf Tiny v0.7 benchmark demonstrated Syntiant NDP120’s outstanding performance, enabled by the Syntiant Core 2’s parallel MAC engines and optimized data path, which avoids wait-states and maintains high utilization.”
Syntiant’s results were performed at two operating points to demonstrate maximum throughput and minimum energy, respectively. With the high-performance setting (1.1V core supply voltage and 98MHz clock frequency), Syntiant’s solution delivers 1.8 ms latency while consuming 49.59 uJ/inference, over 10x faster than any other submitted system. At the low-energy setting (0.9V / 30 MHz), the NDP120 requires 35.29 uJ and 4.3 ms per inference, about 17x more energy efficiency than any other submission. Click here to review the full results of the MLPerf Tiny v0.7 benchmark suite.
Deep Learning Neural Network Architecture
Utilizing the Syntiant Core 2™, a highly flexible, ultra-low-power deep neural network inference engine with a highly configurable audio front-end interface, the NDP120 is easily programmed because of its native support for all major neural network architectures and its direct execution of neural network layers. It can run multiple models concurrently and supports flexible feature extraction and acoustic far-field processing with an included DSP.
Ideal when presence and motion detection are required in ultra-low-power applications, the Core 2 brings always-on neural processing to all types of consumer and industrial edge products, including IP and security cameras, video doorbells, mobile phones, tablets, smart speakers, smart displays, as well as smart home applications and security devices.
Syntiant’s Training Development Kit also provides easy deployment of pre-trained models, such as the benchmark’s reference model. The Syntiant Core 2’s optimized memory access and at-memory compute architecture also provide exceptional efficiency, as demonstrated by the benchmark’s energy results.
The MLPerf benchmarks are full system tests that stress machine learning models, software and hardware and optionally measure power usage. The open-source and peer-reviewed benchmark suites provide a level playing field for competition that drives innovation, performance and energy-efficiency for the entire industry.
More information on MLPerf can be found by visiting https://mlcommons.org.
Contact [email protected] for product related questions.
About Syntiant
Founded in 2017 and headquartered in Irvine, Calif., Syntiant Corp. is a leader in delivering end-to-end deep learning solutions for always-on applications by combining purpose-built silicon with an edge-optimized data platform and training pipeline. Syntiant’s advanced chip solutions merge deep learning with semiconductor design to produce ultra-low-power, high performance, deep neural network processors for edge AI applications across a wide range of consumer and industrial use cases, from earbuds to automobiles. The company is backed by several of the world’s leading strategic and financial investors including Intel Capital, Microsoft’s M12, Applied Ventures, Robert Bosch Venture Capital, the Amazon Alexa Fund and Atlantic Bridge Capital. Syntiant continues to be recognized as an industry leader, being named to Gartner’s Cool Vendors in AI Semiconductors; tinyML Summit’s Best Product of the Year; Fast Company’s prestigious list of the World’s Most Innovative Companies; as well as a CES® Innovation Awards Honoree and CES® Best of Innovation Awards Honoree. More information on the company can be found by visiting www.syntiant.com or by following Syntiant on Twitter @Syntiantcorp.
