AImotive Releases aiWare: The First of its Kind, AI-Optimized Hardware Accelerator For Autonomous Driving
VeriSilicon Signs on as AImotive's First Strategic Partner to utilize aiWare's Hardware IP, which is 20x More Efficient than Leading AI Acceleration Hardware Solutions.
MOUNTAIN VIEW and SANTA CLARA, Calif., June 29, 2017 – AImotive (www.aimotive.com) today announced its much anticipated, AI-optimized hardware IP is available to global chip manufacturers for license. aiWare, built from the ground up for running neural networks, is up to 20 times more power efficient than other leading AI acceleration hardware solutions on the market. VeriSilicon Holdings Co., Ltd. (www.verisilicon.com), a Silicon Platform as a Service (SiPaaS®) company, will be the first to integrate aiWare into a chip design,and the aiWare-based test chips will be fabricated on the GLOBALFOUNDRIES (GF) 22FDX® semiconductor process (www.globalfoundries.com).
AImotive develops a full stack technology solution and the most powerful Artificial Intelligence software for the automotive industry, designed to provide self-driving vehicles better safety, improved comfort and increased productivity. AImotive built aiWare to address the growing need for hardware accelerators optimized specifically for neural networks. aiWare is a power efficient, high-performance neural network hardware IP for automotive embedded solutions, helping chip companies build the optimal hardware to support the performance needs of an AI-driven autonomous driving software. The teams at AImotive and VeriSilicon have a long-standing partnership; aiWare will serve as the core of the demonstration chips designed by VeriSilicon, as the company brings its own silicon expertise to the final product.
“VeriSilicon has a long history of developing cutting edge semiconductor products and we are honored to partner with AImotive on its groundbreaking AI hardware accelerator,” said Wayne Dai, VeriSilicon Chairman, President and CEO. “We are convinced that developing this chip using an advanced process node such as GF 22FDX is a perfect match for the aiWare IP and our leadership position in 22FDX silicon development makes this a natural fit for us.”
The test chips will be manufactured using GF’s 22FDX® process technology, which employs a 22nm FD-SOI transistor architecture to deliver the industry’s best combination of performance, power, and area for a new era of connected intelligent systems.
“The combination of AI with real-time vision processing is a directional shift for autonomous vehicles,” said Alain Mutricy, senior vice president of product management at GF. “With its on-demand performance boost capabilities, our 22FDX process is an ideal platform for power-sensitive AI applications. We are committed to supporting next- generation applications like aiWare that will help shape the AI market of the future, including automotive applications.”
AImotive has been a vocal and ongoing proponent of open collaboration in order to bring autonomous driving to market faster. The company was the first to develop a neural network data exchange standard called NNEF (Neural Network Exchange Format) with the support of the Khronos Group, a global consortium dedicated to standardizing hardware access interfaces. The NNEF standard creates a more accessible and universal AI development ecosystem by making communication simpler and more reliable between AI toolsets and inference engines. The aiWare-based test chip represents the first to meet the new NNEF standard.
“Three components are critical for turning self-driving carsinto a mass product: power-efficient hardware, optimized algorithms and a solid regulatory environment,” said Laszlo Kishonti, founder and CEO of AImotive. “We have gathered experience at delivering the first two elements; now it is up to legislators to catch up with the industry.”
AImotive, VeriSilicon and GF plan to debut the new aiWare-based chip in early 2018. In the meantime, AImotive will offer an FPGA evaluation kit for global semiconductor companies that further demonstrates the value of aiWare.