Featuring speech inference at 280 Watts, NDP115 is looking to make a home in wearables and smart appliances.
Syntiant NDP115 Neural Decision Processor
Syntiant’s NDP115 Neural Decision Processor™, the latest addition to the company’s family of special-purpose silicon, was unveiled on Jan 4 by Syntiant Corp., a deep learning solutions provider for edge AI applications. It was built using the company’s Syntiant Core 2™ inference engine, which can run multiple neural network loads at less than 1mW.
Despite being small, the NDP115 can deliver highly accurate, cloud-free audio and sensor AI processing for a broad range of edge products, from smart homes, like remote controls, to consumer electronics, such as hearables and wearables, to industrial solutions for predictive maintenance. It is available in a 2.1 mm x 2.1 mm 25-ball WLBGA package (0.4 mm ball pitch).
CEO of Syntiant, Kurt Busch, said, “ The NDP115 offers the multi-modal functionality of our Core 2 inference engine in a compact, cost- and power-efficient solution for ultra-power and size-constrained applications.” Furthermore, he added, “combined with our machine learning software models, the purpose-built NDP115 enables developers to easily deploy full audio and sensor processing solutions that address all kinds of consumer and commercial use cases, from home security to industrial IoT.”
The NDP115 is built to natively operate several deep neural networks on various architectures, including CNNs, RNNs, and fully connected networks, and is capable of making speech inferences at 280 microwatts. Sensor fusion, tilt, magnetic field, and pressure are all supported by the NDP115 via I2C and pulse density modulation (PDM). It is ideal for close-talk, far-field, keyword speech, and audio event classification applications.
Key Product Features
The incredible features of the product include the following;
- Embedded Arm Cortex M0 microcontroller.
- Syntiant Core 2 neural processor.
- Assisting concurrent neural networks comprising multi-sensor fusion.
- Embedded user-programmable HiFi-3 DSP with 64KB 64-bit instruction RAM and 144KB 64-bit data RAM.
- Direct input of PCM audio and aid for up to five audio streams.
- Flexible clock generation.
- Onboard firmware decryption and authentication
- 25-ball WLBGA package (0.4mm ball pitch)
- Upto 13 GPIO pins
- 12C target modes and controller for integration and sensor control
Pricing
The Syntiant NDP115 is now available for delivery in production volumes. The cost per unit for 10Ku batches is $3.25.
CES 2023
At CES 2023, Syntiant will showcase its end-to-end deep learning solutions, including the NDP115 Neural Decision Processor at the Venetian Hotel, for always-on vision, audio, and sensing applications (Room 108, 36th floor). To schedule a demonstration of the company’s technology usage in battery management systems, smart homes, event detection devices, and teleconferencing solutions, among other use cases, go to www.syntiant.com/ces.
About Syntiant
Syntiant Corp., established in 2017 and with its headquarters in Irvine, California, is a pioneer in providing end-to-end deep learning solutions for edge deployment. Global deployments of the company’s custom-built silicon and hardware-independent models power cutting-edge AI voice, audio, sensor, and vision applications in various consumer and commercial use cases, from earbuds to automobiles.
Deep learning and semiconductor design are combined to create ultra-low-power, high-performance deep neural network processors by Syntiant. Additionally, Syntiant offers compute-efficient software solutions with patented model designs and hardware-specific optimizations, enabling industry-leading inference speed and a reduced memory footprint on various processors.
Several top strategic and financial investors worldwide, including Robert Bosch Venture Capital, Atlantic Bridge Capital, Intel Capital, Microsoft’s M12, Applied Ventures, and the Amazon Alexa Fund, support the company. Visit Syntiant Website to learn more about the business, or follow Syntiant on Twitter at Syntiantcorp or LinkedIn.
