Ambient Scientific has announced that its first AI System on Chip (SoC) – the GPX10 – is now in volume production. Designed for running AI applications without Cloud dependency on portable, battery-powered devices, the GPX10 is manufactured using TSMC’s 40 nm technology. This product comes equipped with a comprehensive development platform, including a Software Development Kit and Neural Network Compiler, to enable On-Device/IoT AI Applications. The Evaluation kit for GPX10 will be available for commercial purchase starting from Q1 2024.
The GPX10 AI SoC, which is smaller than a cent coin, delivers exceptional AI throughput for on-device and edge AI applications. Created using the breakthrough DigAn technology, GPX10 combines high-speed digital and analog circuit technologies to provide leading AI throughput at minimal power consumption, setting a new standard for efficiency. The GPX10 offers significant advantages for edge AI, such as:
- Performance Efficiency exceeding 8 TOPS/W for on-device AI applications at 40nm.
- Power consumption under 100 µW for always-on sensor fusion applications, enabling operation on coin cell batteries.
- A compact form factor, smaller than a fingernail, suitable for tiny portable devices.
- Enhanced privacy and security due to reduced data transfer over the internet.
GPX10 is fully programmable and, coupled with Ambient Scientific’s software toolchain, including an AI compiler and development libraries, enables battery-powered applications across various sectors like automotive, healthcare, consumer wearables, and industrial applications.
Satish Kutty M, Head of Strategy and Business Development at Ambient Scientific, said, “We’re excited to be already partnering with customers to deliver AI applications powered by GPX-10. These include a driver monitoring system, ECG wearables, and always-on voice-based distress systems.” He anticipates the industry’s innovative use of the GPX-10 AI processor with Ambient Scientific’s edge AI development platform.
Ambient Scientific is on a mission to democratise sustainable AI. “To bring AI to the edge, it was essential to liberate it from cloud and internet dependencies, which required a complete rethinking of conventional computing architecture,” stated GP Singh, Founder and CEO of Ambient Scientific.
The team achieved this through innovations at every level, from circuit technologies to a new Instruction Set Architecture (ISA), including:
- The DigAn Matrix computer, combining digital and analog circuit technologies for high-performance, low-power matrix computing.
- 3D memory with in-memory computing to assist the DigAn matrix computer in efficient operand distribution.
- Multi-channel Ultra Low power Sensor ADC for connecting multiple analog sensors simultaneously.
- Always-on Sensor Fusion DMA for microwatt-level intelligent sensor AI applications on multiple sensors data fusion.
GP Singh, Founder and CEO of Ambient Scientific, added, “We have blended digital and analog computing to harness the scalability of digital and the efficiency of analog. Our processors are manufactured and can be programmed like any digital chip while offering the power and area efficiency of analog.”
Ambient’s MX8 AI core, based on this technology, is highly scalable across any number of cores and process nodes, leading to a robust range of processors. These range from GPX-1, designed for integrating with MEMs sensors, to GPX-64, a chip focused on high-speed video AI applications.
With the GPX10 entering volume production and set for market release in Q1 2024, Ambient Scientific is advancing its development of new products to achieve its mission of democratising sustainable AI for the betterment of human lives.
There’s plenty of other editorial on our sister site, Electronic Specifier! Or you can always join in the conversation by commenting below or visiting our LinkedIn page.
