The Neural Decision Processor claims to be a powerful chip for imaging, speech, and sensor applications, featuring low power consumption and integrated processing capabilities.
Syntiant Corp. has launched the NDP250 Neural Decision Processor (NDP), which the company claims to be the most powerful chip with 5x the tensor throughput compared to its previous generation. Built using the Syntiant Core 3 programmable deep learning architecture, the NDP250 is designed for imaging, speech, and sensor applications in power-constrained devices, such as person detection, object classification, automatic speech recognition (ASR), text-to-speech (TTS), and motion tracking.
Integrating ASR and TTS into devices running on the NDP creates a pathway to utilizing LLMs as the voice interface. The chip’s processing capabilities enable these features to understand and produce human language, improving interactions.
The chip features advanced image capabilities, making it suitable for ultra-low power video applications in automotive security, battery-powered cameras, and video doorbells. Operating always-on image recognition at under 30mW offers several benefits:
It reduces power consumption by locally processing device data, extending battery life and optimizing resource usage. It lowers latency as data doesn’t need to travel to and from a remote server, resulting in faster response times crucial for real-time applications and enhancing customer satisfaction. Additionally, it improves privacy by processing sensitive data locally, minimizing the need to transmit information over networks where it could be vulnerable to breaches or interception. Moreover, it reduces cloud costs, as transferring and processing less data in the cloud lowers infrastructure expenses for businesses deploying edge AI solutions.
The chip is equipped with an Arm Cortex M0 processor and a HiFi 3 DSP to support feature extraction and signal processing for image and voice enhancements. It also features an integrated power management unit that allows single power rail operation. The integrated phase-locked Loop (PLL) further optimizes system cost and size.
Furthermore, the NDP can process multiple heterogeneous networks concurrently, supporting convolution neural networks, including 1D, 2D, and depth-wise, fully connected networks, and recurrent neural networks, including LSTM (long short-term memory) and GRU (gated recurrent unit).
For more information, click here.
