Syntiant Corp. is expanding its line of special purpose artificial intelligence (AI) silicon with a neural decision processor designed for a wide variety of edge products.
Called the NDP115, the processor is built using Syntiant’s Core 2 inference engine that can run multiple neural network loads at under 1 mW. The device comes in a 2.1 mm x 2.1 mm 25 wafer level ball grid array (WLBGA) package but delivers enough performance for cloud-free audio and sensor AI processing in applications such as hearables and wearables, smart home remotes, industrial predictive maintenance and more.
The neural decision processors can run speech inferences at 280 microwatts and can natively run multiple deep neural networks on a variety of architectures such as convolutional neural networks (CNNs), recurrent neural networks (RNNs) and fully connected networks. It could be used for close-talk, far-field, keyword speech and audio event classification applications.
Other features include:
- Support for concurrent neural networks, including multi-sensor fusion
- Embedded user programmable HiFi-3 DSP with 144 KB 64 bit data random access memory (RAM) and 64 KB 64 bit instruction RAM
- Embedded arm cortex M0 microcontroller
- Direct input of pulse code modulation (PCM) audio and support for up to five audio streams
- I2 C controller and target modes for sensor control and integration
- Up to 13 general-purpose input/output (GPIO) pins
- Flexible clock generation
- Onboard firmware decryption and authentication
