The company originally announced the design back in November 2016 (see IoT processor beats Cortex-M, claims startup) but then had to wait until December 2017 before it could tape out. Martin Croome, vice president of business development, said it took longer than expected to raise funds but the time was also used to make some improvements to the design
GreenWaves eventually announced the raising of €3.1 million of funding (about $3.8 million) in August 2017.
The GAP8 is based on the RISC-V opens-source hardware PULP core developed at the Universities of Bologna and ETF Zurich (see Swiss open-source processor core ready for IoT ). It includes eight such cores plus a ninth as a controller for a microcontroller section. There is also a hardware convolution engine (HWCE) to accelerate neural network operations.
The chip has been designed in the 55LP 55nm CMOS process from foundry TSMC. This operates at 1.2V nominal in the core and has an I/O that operatest at 1.8 to 3.3V. The chip is intended to operate at clock frequencies up to 175MHz providing significant performance from its 8-core cluster. The design it makes use of extensions to the RISC-V instruction set to help with signal processing and boosting of convolutional neural networking performance.
The GAP8 is decribed as an IoT application processor and its intended to operate at the edge of the network autonomously to capture, analyze, classify and act on the fusion of data sources such as images, sounds or vibrations. GAP8 is optimized to execute image and audio algorithms including convolutional neural network (CNN) inference, with extreme energy efficiency. A separate core, within an independent voltage and frequency domain, takes care of communication, control and information pre-analysis. This allows industrial and consumer product manufacturers to integrate artificial intelligence and advanced classification into new classes of wireless sensing devices for IoT applications including image recognition, counting people and objects, machine health monitoring, home security, speech recognition, consumer robotics, wearables and smart toys.
Next: Production volumes and development board
Engineering prototypes and development boards are due to be available in April 2018 with production set to follow in 4Q18, Croome said.
One conventional microcontroller offering missing from the GAP8 is an analog-to-digital converter. Croome said the breadth of applications meant that selecting one particular or multiple ADCs for inclusion was not appropriate. However, the GAP8 can work with sensors with digital outputs or an off-chip ADC to provide a sensor fusion and analysis node.
Use cases for the GAP8 include: always-on face detection consuming a few milliwatts of power; indoor people counting/presence detection; machine vision and voice control solutions for consumer robotics; single-chip processing for four-microphone voice capture and ten-word speaker independent keyword spotting.
For most developers, GAP8 is programmed like a microcontroller. When compute-intense tasks need to be launched they are off-loaded to the cluster through APIs included in the GAP8 software development kit. It is also possible to take trained CNNs described with an AI framework to be optimized for and ported onto the GAP8.
“GAP8 differs radically from the flurry of other AI focused processors, which target either far more complex applications that can’t be battery powered, or much narrower application spaces. Instead, GAP8 is precisely positioned at the crossroads of the AI, IoT and MCU worlds. The autonomous operation enabled by GAP8 dramatically reduces the deployment and operating costs of image, sound and vibration sensing devices, enabling an unprecedented scale of adoption,” said Loïc Liétar, co-founder and CEO of GreenWaves Technologies, in a statement.
The chip is set to be priced at €5 in 100,000 unit volumes.
Related links and articles: