Ultra-Low Power On-Device Digital-Analog Artificial Intelligence
Ambient Scientific Inc. today unveiled industry's first digital analog (DigAn™) artificial intelligence (AI) technology with demonstration of its first DigAn™ AI processor chip at the Edge AI Summit. DigAn™ AI is an advanced version of analog AI technology that provides ultra-high performance and efficiency of analog AI with implementation and commercial advantages of digital chips.
The company also announced Industry's lowest power on-device AI solution for ‘wake on word' applications.
Ambient Scientific's DigAn™ AI processor is built with multiple fundamental grass root level innovations. These include Analog Matrix Compute Engine using DigAn™ technology, that performs blazingly fast arithmetic computations with un-precedented power efficiency, ultra-low power embedded SRAM for critical storage, and ultra-low power on-chip interconnect technology that enables massive parallel processing.
Ambient's DigAn™ AI processor computes an entire layer of neurons in a single clock cycle using custom analog techniques, a task that takes several hundreds or even thousands of clock cycles in traditional implementations. However, DigAn™ technology uses standard digital transistors to implement its analog functions.
Experts agree that Analog is a great fit for AI computing. However, prior implementations had severe limitations in accuracy and computing range - two important parameters for optimal AI solution. Ambient has eliminated these limitations. While prior implementations have struggled to achieve 8-bit accuracy, Ambient's DigAn™ technology can achieve 32-bit accuracy. This is 16 million times more accurate in computer language.
Ambient's first solution is targeted towards Industry's lowest power On-Device ‘always-on' AI for ‘wake on word' and anomaly detection applications. It uses only 80μW or less for AI computing, which is 200x lower power than industry standard and 2x to 3x lower than recent advancements.
In addition, Ambient has invented important components such as ultra-low power analog-to-digital converter (ADC) to ingest sensor data and ultra-low power clock generator (LPO) to enable chip operation without external host.
"AI computing is different. It requires fundamental innovations grounds up. Ambient has invested heavily in inventing optimal circuit components for AI computing. These enable a revolutionary AI Processor architecture that provides ultra-high performance at extremely low power consumption. This processor is easy to program, flexible to use and scalable for growth", says Gajendra Prasad Singh, CEO.