Guangzhou, China, September 21, 2020 – Guangdong Gaoyun Semiconductor Technology Co., Ltd. (hereinafter referred to as “Gaoyun semiconductor”) today released the latest version of goaI ™ Machine learning platform, which provides SDK and accelerator, builds convolutional neural network on high cloud FPGA to perform machine learning of edge reasoning. Gao Yun goaI ™ 2.0 can be directly integrated into tensorflow and tensorflow Lite machine learning platforms for high cloud gw1nsr4p µ SOC FPGAs are optimized and accelerators are provided to share computing intensive functions from microcontrollers embedded in high cloud FPGAs, thus improving performance by 80 times.
With the rapid development of machine learning, in order to achieve better standardization, reliability and development simplicity, the framework, platform, model and data set are becoming consistent. Tensorflow has become one of these platforms and supports embedded SOC and microcontroller. GoAI ™ The new features of 2.0 make it easy for customers to use tensorflow which supports high cloud embedded FPGA.
“GoAI ™ 2.0 has several important updates on embedded FPGAs that deploy machine learning models to focus on the edge. ” Grant Jennings, international marketing director of Gaoyun semiconductor, said“ We can use goaI ™ 2.0 deploy the machine learning model of mobilenet to our gw1nsr4p µ SOC FPGA. Gw1nsr4p is very suitable to use tensorflow Lite for microcontrollers to perform tinyml reasoning at the edge, because it includes an arm Cortex-M3 hard core microcontroller for directly porting models and controlling goaI accelerators; The 4.6k look-up table constructed by FPGA is used to instantiate goaI ™ 2.0 accelerator and connected sensor input; It also includes an additional 8MB of SRAM for layer storage, all of which are contained in a low-cost 6x6mm qfn48 package. Our goaI ™ The 2.0 SDK enables our customers to quickly and easily deploy tensorflow to FPGA.
Using goaI ™ 2.0, without FPGA RTL or microprocessor C / C + + programming. GoAI ™ 2.0 SDK automatically generates C / C + + code to drive accelerator from arm Cortex-M processor. GoAI ™ 2.0 accelerators are provided as FPGA IP, but are also included in partially pre generated FPGA bitstreams with various types of sensor inputs. GoAI ™ The design of 2.0 accelerator architecture makes the processor only need to update a register mapping in each model layer, and does not need RTL changes to deploy or change the machine learning model used.
Although machine learning may have an obvious learning curve, goaI ™ 2.0 provides an introduction to several different types of sensor inputs, such as cameras, microphones and accelerators. The reference design includes data inference to predict sine wave output based on input value, audio phrase detection to infer “yes” or “no” from microphone input, and human detection to infer whether someone exists in camera view. More reference designs are still under development, including gesture detection using accelerometers to detect the shapes drawn in the air when the user holds the development board. Developers can use three goaI ™ One of the embedded development kits starts to use goaI ™ 2.0。 The development kit includes high cloud FPGAs of various densities, sensors for running reference designs, and other peripherals, such as HDMI inputs and outputs for video demonstrations.
“Marginal machine learning reasoning has become the mainstream of many market applications, including consumer, industrial and medical applications.” Scott Casper, sales director of Gaoyun semiconductor in America, said“ GoAI ™ The 2.0 platform enables embedded system engineers to easily add these advanced functions to applications at a cost-effective price. We provide a lot of resources, such as reference platforms, development kits and on-site support, to make it easy for engineers to get started immediately. “
“In the coming AI Expo Korea exhibition, Gao Yun will show goaI ™ 2.0 platform. ” Xie Zhaojian, Asia Pacific Sales Director and general manager of Gaoyun semiconductor in Hong Kong, said, “we will show the goaI 2.0 platform for Korean innovation and leading AI technology for the first time in AI Expo Korea. Gao Yun goaI ™ 2.0 provides the most cost-effective platform for the edge AI market, and our personnel detection scheme has been selected as the next generation platform by Korean customers. Compared with their existing solutions, our platform greatly reduces power consumption and design complexity. “
Gao Yun’s goaI ™ The 2.0 solution will be displayed at the D Pavilion of Coex in Seoul, South Korea from October 27 to October 29, 2020.