In-Sensor Analog Neural Network Framework for Analog to Information Conversion

Analog sensors are expected to produce 1020 bits of data per second by 2030, and data transmission for such high volumes of data, especially for wireless transmission, is extremely energy expensive. Digitizing the analog sensor data for transmission to remote location (e.g., cloud server) for processing creates an enormous amount of data that is predicted to grow at exponential rates. Improvements in the transmitter chain, such as higher energy-efficiency analog-to-digital converters (ADCs) or data compression techniques (compressive sensing), are not enough to result in orders-of-magnitude reduction in transmission volume and energy. Instead, addressing the data deluge problem requires re-imagining the entire signal acquisition chain and possibly drawing inspiration from human brain to efficiently extract information from the large volume of sensor data. General statement of how this problem is being addressed. Alphacore proposes embedding artificial intelligence (AI) locally on-sensor to intelligently determine/extract information from sensor signals and transmit only meaningful selected segments to achieve reduction in sensor data transmission volume and energy. The driving motivation behind on-sensor AI is to achieve sensor data compression ratios of 105:1 that is only possible through intelligent algorithms that can detect subtle and ‘rare’ pattern/information hidden in large volumes of sensor data. In addition, local AI capability lowers latency compared to remote (cloud-based) processing and is critical for applications that need real-time monitoring, such as for predictive maintenance in industrial manufacturing applications where in-sensor AI models can continuously monitor sensor data and look for early signs of malfunction. What is to be done in Phase I? During Phase I, Alphacore will meet the following objectives: Objective 1: Design of reservoir-computing network for sensor data classification Objective 2: Design in-memory computing circuit for read-out layer of reservoir-computer Objective 3: Analyze the performance improvements (in terms of data compression and energy reduction) of in-sensor AI across applications Commercial Applications and Other Benefits (limited to the space provided). Alphacore’s solution will enable sensors to determine and extract selected segments of the data signals, and transmit only the relevant data, achieving reduction in the data transmission volume, and therefore, energy consumption. Given the ubiquitous importance of saving energy, and the continuous demand to achieve as little power consumption as possible without sacrificing system performance, the market opportunity for this innovation is expected to be massive. Applications that will benefit include industrial automation and manufacturing, power and utilities, automotive applications, smart homes and cities, and more Internet-of-Things related applications. The main benefit for all these industries will be increased efficiency and reduced costs in data transmission and communication within the sensor networks, while progressing towards decarbonization and reducing greenhouse gas emissions.