AMMTO Semiconductor Workshop 3

Workshop 3

Analog & Neuromorphic Computing

August 11-13, 2021

The U.S. Department of Energy’s (DOE’s) Office of Energy Efficiency and Renewable Energy (EERE) Advanced Manufacturing Office (AMO) hosted the third in its series of semiconductor R&D workshops on August 11-13, 2021. This workshop focused on RDD&D opportunities to improve device and manufacturing capabilities for analog and neuromorphic computing approaches that have the potential to increase the energy efficiency of microelectronic systems.

Digital storage and computation underpin the success of nearly every modern industry. Since 2010, semiconductor energy use has doubled every 3 years. By 2030, semiconductors could consume nearly 20% of planetary energy production. Hence innovation in semiconductor energy efficiency is essential for the economy, jobs, and to address the climate crisis.

In particular, advances in modeling, simulation, artificial intelligence, machine learning, and other analysis and discovery techniques are driving innovations in manufacturing processes and approaches, reducing energy consumption, and limiting GHG emissions. To permit continued US participation and manufacturing growth in these application areas, the underlying microelectronic devices need to continue to improve as well.

As improvements in the size and performance of conventional electronic devices reach their physical limit, industries are pursuing multiple innovation pathways to prevent stagnation. There are many application areas for computing (such as communication and sensing) and approaches to computing that use analog hardware and techniques to increase speed and efficiency.

Performing computation on analog signals (e.g., sensor input and RF signals) in the analog domain (before conversion to digital memory) maintains signal fidelity and in many cases can optimize computing efficiency in ways digital computation can’t. Similarly, analog signal processing can greatly improve the bandwidth of wireless communication. Though basic analog computing has a long history, new opportunities exist.

Communication and sensing are two domains where advanced analog hardware can drastically improve energy efficiency of microelectronic systems. Reducing data generation and data transmission of integrated sensor systems were two of the major conclusions of the first workshop due to their significant energy penalty. In addition, analog devices that utilize novel materials or architectures can provide significant energy savings while maintaining performance.

Neuromorphic computing has emerged as a major driver of machine-learning (ML) capabilities, including advanced process optimization, object recognition, and speech recognition due to the 1000x improvements in efficiency. Advanced devices for neuromorphic computing offer a hardware solution to the limitations and inefficiencies of software and conventional CMOS-based neuromorphic methodologies.

Neuromorphic hardware has shown promise to drastically reduce the computational energy use and data generation compared with conventional approaches. During the workshop, representatives from the research community discussed the latest advances in analog and neuromorphic computing hardware technologies, which industries and applications are best suited to benefit from them, and the engineering barriers and challenges that slow or prevent scale-up and potential integration with current semiconductor technologies.

Workshop 3 Summary Workshop 3 Read-Ahead Documents Workshop Series Workshop 3 Report