LAS VEGAS, N.V., - Consumer Electronics Show - Jan. 12, 2018 - Today Hua Xia GPT (hxGPT), China’s leading licensor of customizable processor IP cores, announced a scalable, programmable AI accelerator family available for licensing at CES 2018. Solutions scaling from handset to edge processing support a wide range of applications including the Internet of Things (IoT), Advanced Driver Assistance Systems (ADAS), Self-driving Cars, Robotics, Surveillance, and Machine Vision.
The “Song” AI core is a highly configurable IP block that provides for efficient implementations of CNN’s. Running at up to 1GHz in 28nm technology, each accelerator core contains 288 MAC units scalable to beyond 2 TFLOPS. Performance, power, and size can be optimized efficiently supporting both low power embedded IoT devices, handsets, and edge processing designs. Accuracy can be optimized using built-in 16-bit integer or floating point operations. Support for compressed 8-bit data is provided. Extended precision intermediate results reduce errors in both regression and classification workloads.
Targeting inference applications, the AI accelerator supports multi-dimensional tensor processing. The highly optimized design provides direct support for convolution, pooling, dropout, padding, and programmable activation functions. A library of popular CNN networks is provided with generalized framework support (e.g. Tensorflow, Caffee, etc.) in development.
Previously hxGPT announced availability and licensing of the Changcheng out-of-order superscalar CPU. Validated in TSMC’s 28nm process, the core is configurable providing for multiple performance levels based on customer requirements. Memories, caches, and out-of-order features (issue window, ROB, rename, etc.) can all be customized. Depending upon technology selection the core can run up to 3GHz at very low power.
Bulding on hxGPT’s expertise in heterogeneous computing, the Song AI core easily integrates with hxGPT’s Changcheng CPU for ease of programming. Both are programmed using the same familiar processor programming environment.
WuxiDSP, hxGPT’s customer of both the Changcheng core and Song AI core noted “hxGPT’s cores are highly optimized and power efficient. Our first silicon using both of these cores is scheduled to sample later this year targeting edge IoT processing,” said Xiaodong Zhang, Chief Engineer of WuxiDSP.
hxGPT is the only IC design house in China with a portfolio of proprietary intellectual property for CPU, DSP, GPU and AI accelerators targeted at high-end SoC designs for heterogeneous computing. The company has 8 issued patents and more than 90 pending world-wide patents covering the core architecture and implementations of the next generation processor design. hxGPT is a global corporation with operations in New York, Beijing, and Shanghai.
Engaging the global market, hxGPT pursues excellence by innovating at all levels of design from algorithms to semiconductors. Providing power and performance optimized solutions requires heterogeneous cores capable of efficiently processing application specific workloads. Unique to hxGPT is a common programming environment with integrated debugging and profiling.
hxGPT’s core business is licensing of proprietary IP cores and helping customers to integrate them into customized SoC solutions. Applications adopting these cores using AI accelerators include Assisted and Self-driving Cars, Robotics, Surveillance, and Machine Vision.