Syntiant Unveils NDP250: A Breakthrough in Microwatt AI Processing for Always-On Applications

Syntiant, a leading provider of on-device deep learning solutions, has introduced its latest innovation: the NDP250 Neural Decision Processor. Boasting an impressive 30 giga-operations per second (GOPS) performance, the NDP250 sets a new standard for efficiency in AI processing. This third-generation chip represents a significant leap forward, offering five times the speed of its predecessor, the Core 2.

“Our NDP250 builds on two generations of neural network architectures to deliver 30 GOPS, making it our fastest, highest-performing chip yet,”

says Syntiant CEO Kurt Busch, who unveiled the chip at Embedded World.

Designed for always-on voice and vision applications, the NDP250 can power various tasks, from speech synthesis to image recognition. Its advanced architecture, including the innovative Core 3 engine, enables seamless multitasking while operating within a microwatt power envelope.

Busch further emphasizes the versatility and efficiency of the NDP250, stating, “Compatible with a host of architectures while running multiple different layers simultaneously at significantly less power than existing solutions, the NDP250 with our new Core 3 engine is the ideal real-time speech interface for large language models. It can bring powerful AI to battery-powered, always-on vision applications in automotive security, appliances, cameras, smart displays, and video doorbells.” He highlights the transformative impact of the NDP250, noting that applications that once demanded power measured in watts can now operate with power measured in microwatts.

The NDP250 adopts a familiar approach seen in the company’s previous “Neural Decision Processors,” but with the introduction of the new third-generation Syntiant Core 3 architecture. This upgrade brings significant enhancements, with the company claiming a fivefold increase in performance compared to its second-generation counterparts, all without increasing power requirements. The chip’s architecture includes an Arm Cortex-M0 processor alongside Syntiant’s Core 3 and a HiFi3 programmable digital signal processor (DSP). Additionally, it features a dual 11-wire direct image interface, dual PDM digital microphone interface, I2S with PCM support, and quad-SPI and dual-I2C buses, complemented by 58 general-purpose input/output (GPIO) pins.

According to Syntiant, the NDP250 excels in handling multiple heterogeneous networks concurrently, supporting over six million total neural parameters when operating in eight-bit precision mode. It can accommodate various neural network architectures, including 1D, 2D, and depthwise convolutional neural networks (CNNs), fully-connected networks, and recurrent neural networks such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks. Besides real-time speech recognition and voice synthesis tasks, the chip is capable of efficiently running always-on image recognition models while consuming less than 30mW of power.

Syntiant is showcasing the NDP250 at its booth during Embedded World 2024, located at Hall 2 Booth #2-238. The company has confirmed that the chip is currently available for sampling; however, it has not provided a timeline for general availability.

More information on Syntiant’s chips is available on the company’s website.

Please follow and like us: