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

The Department of Energy Advanced Manufacturing Office (AMO) has a critical need for advanced semiconductor technologies that can be applied to next-generation manufacturing needs to help bring U.S semiconductor manufacturing leadership into the forefront of global supply and improve manufacturer decarbonization abilities with more efficient operations and better monitoring of greenhouse gas emissions. Pivotal analog electronics determined to achieve widespread benefits from use of sensors include 1) Analog electronics connected directly to the sensor, in front of the ADC conversion, to reduce the ‘data’ deluge’ and 2) analog electronics for robust wireless communication (5th or 6th generation) for fast, efficient control and/or data collection. For this critical ‘data deluge’ need, Alphacore has developed an ‘In-Sensor Analog Neural Network Framework for Analog to Information Conversion” (ISANNAC) technology with 105 :1 compression factor to reduce total data processed while its on-chip ultralow power ADC-based digitization further reduces transmission energy. Digitizing the analog sensor data for transmission to remote location (e.g., cloud server) for processing creates an enormous amount of data. The two key issues with this level of data generation are: a) processing this large volume of data is beyond human capacity which is estimated to be 1017 bits/second; b) transmitting all the data from sensors to remote locations is prohibitively energy expensive. To achieve this, Alphacore worked extensively on two major innovations that realized this concept: 1) to use reservoir-computing (RC) paradigm that nonlinearly projects the analog sensor data into a different hyperplane, allowing for easy separation of the distinct classes in the input signal by a read-out layer (Error! Reference source not found.) and 2) a new ultra-low-power ADC for digitizing the sensor analog signals. Alphacore had set the goal of meeting the following three milestones: Milestone 1: Post-layout design of the reservoir-computer circuit has been completed and evaluated; Milestone 2: Layout of complete in-sensor classifier and characterization; Milestone 3: Tradeoff analysis completed, Phase II schedule planned carefully, and Phase I report submitted to DOE. These milestones were successfully met. In addition, during the Phase I study, Alphacore achieved two additional major technology milestones: Milestone 4: Completed fj/conversion scale SAR ADC; Milestone 5: Completed integration of RC circuit and SAR ADC and sent test chip to fabrication. The additional milestones were to complete the design of the novel “disruptive” low-power ADC and to design, layout and send to the fab the first test chip that includes the RC circuit integrated with the ADC. Phase II follow-on program is hereby proposed where our work is to complete the development of a system that implements this framework as a full silicon integrated circuit. This includes design, layout, verification, fabrication, and functional test of the prototype IC. The on-chip innovation will be tested with a high-bandwidth sensor to benchmark its performance. Reducing manufacturing energy intensity with clean energy methods that implement decarbonization, integrating energy efficient sensors that can function in harsh environments (including hostile temperature, radiation and chemical), implementing methods for efficient, co-integrated sensor node data acquisition and processing, and developing methods for efficient local computing and/or data transmission to realize >10x reduction in energy consumption of sensor data transmissions has been recognized as a critical need, therefore an AI-based data compressor that reduces total data processed by 105 :1 with an ultra-low power ADC digitizer to further reduce transmission energy will be a competitive disrupter. Alphacore’s primary target market for this ISANNAC technology are those using industrial manufacturing imaging systems, RADAR and LIDAR systems, scientific research groups (such as High Energy Physics (HEP) and Nuclear Physics (NP), and medical imaging systems.