Neural network accelerator adds AI to edge devices

Two neural network cores from Imagination Technologies are designed to enable high-performance computation of neural networks at very low power consumption in minimal silicon area.  The AX2185 and AX2145 cores are based on Imagination’s neural network accelerator (NNA) architecture, PowerVR Series2NX, which enables ‘smartness’ to move from the cloud into edge devices, says the company. This enables greater efficiency and real-time responsiveness in consumer devices.

The Series2NX AX2185 targets the high-end smartphone, smart surveillance, and automotive markets, where neural network acceleration has a significant impact in areas such as image categorisation and driver assistance systems. There are eight full-width compute engines in the AX2185 core, which provides 2,048 MACs/clock (4.1Tera operations per second). According to Imagination, this represents the highest performance per mm2 in the market today.

Optimised for cost-sensitive devices, the AX2145 core targets mid-range smartphone, DTV/set-top box, smart camera and consumer security markets, all of which use neural network acceleration for various tasks such as image management and vision-based applications. The PowerVR AX2145 streamlined architecture delivers performance-efficient neural network inferencing for low bandwidth systems, and represents a cost-effective solution for OEMs and ODMs working with limited silicon area budgets, adds Imagination.

Both cores support the Android Neural Networks API (NNAPI), used by developers to bring neural network capabilities to Android-based mobile devices.

The PowerVR Series2NX architecture was designed from the ground-up to provide hardware acceleration for efficient neural network inference in mobile and embedded platforms. Its flexible bit-depth support on a per-layer basis for weights and data means PowerVR Series2NX can maintain high inference accuracy while reducing bandwidth/power requirements. It is the only solution supporting bit-depths from 16- to four-bit, resulting in higher performance at lower bandwidth and power, claims Imagination.

To accompany the PowerVR Series2NX cores, Imagination provides a comprehensive set of tools to streamline AI application development and deployment and, debugging and analysis. Machine learning frameworks are supported using our Network Developer Kit (NDK).

The AX2185 core has already been delivered to lead customers and both cores are available for licensing now.

http://www.imgtec.com