With remarkable progress in transistor technology, circuits can now be built in all manner of embedded substrates, e.g., flexible, stretchable, conformal and impact-resistant and large-area formats. Such technologies will greatly expand the application space for microelectronics, including radiation detection, health diagnostics, drug-delivery, distributed sensing, information display, food security, identification tagging, inventory tracking, robotics, and human-machine interfacing. The aim of this project is to provide a proof-of-concept, end-to-end validation of the paradigm of stochastic bit stream computation, applying it to a promising new technology, namely, printed flexible electronics. Compared to the current CMOS technology, flexible electronic systems have severe limitations on the number of devices that can be built on a given surface, and thus, it becomes critical to perform computations with a very small number of transistors. The goal of this project is to use probabilities to represent numbers and employ a novel circuit design paradigm to significantly reduce the number of transistors used in flexible electronics. The education plan accompanying the project will promote probabilistic computing in the classroom, training students to design end-to-end systems by applying multiple technologies. Flexible electronics and robotics will be leveraged to mentor K-12 students and promote engineering education amongst minority and female students by taking advantage of existing institutional outreach programs such as Lotus Stokes Alliance for Minority Participation and Northstar STEM Alliance.
This proposal seeks to apply the paradigm of stochastic bit stream computation to the design challenge of printed electronics. In this paradigm, circuits compute on random bit streams with signal values encoded by the probability of obtaining a "one" versus a "zero" in the streams. Because the bit stream representation is uniform, with all bits weighted equally, circuits designed this way are highly tolerant of soft errors. More importantly, complex operations can be performed with very simple logic. For instance, multiplication can be performed with a single AND gate. In general, the methodology provides significant reduction in transistor counts. The test platform will consist of a low voltage, flexible pressure sensor array with embedded stochastic computational elements to act as an electronic skin for a robot's foot. Broadly, the project will develop the following tasks as part of the test platform: synthesis methodologies; architectures and applications; input/output interfacing; and developing new low temperature, additive manufacturing approaches to printed electronics that will decrease device footprints, thereby simultaneously increasing the device count per area and increasing bandwidth.
