Bia: A Co-Design Methodology to Transform Materials and Computer Architecture Research for Energy Efficiency

The Bia project tackles a major challenge facing scientific research: how to improve both performance and energy efficiency in systems that gather and process enormous volumes of data. To address this, Bia uses a ​“co-design” approach. This means experts in different areas — such as materials science, device design and computer architecture — work together from the start to ensure new ideas fit together smoothly and effectively.

At the hardware level, the Bia team is developing neuromorphic devices, which are electronic components modeled after the way the human brain processes information. One example is the memtransistor, a low-power version of a traditional transistor that stores data much like a biological synapse. Researchers are also exploring van der Waals heterojunctions, where ultra-thin layers of different materials are stacked to form next-generation circuits. 

On the computing side, the project is developing simulation tools, novel accelerator designs and software/hardware experiment frameworks that enable science teams to rapidly create and evaluate new experiments, increase their capabilities and achieve results in shorter time.  These frameworks can process massive data sets in real time, and the accelerators will dramatically increase speed, computing and scientific capabilities. These advances will accelerate experiments in high energy physics and X-ray science. Such fields generate tremendous amounts of data when smashing particles together or imaging materials at the atomic scale. 

Bia: Need for a Co-Design Methodology 

Focus on energy efficiency 

• Focus on two design problems
• Future detectors will require at least three orders of magnitude increase in data rates next two decades
• Two vertical design problems
  • Data reduction and transport for HEP and XRS detectors
  • Feature extraction, high level analysis and feedback control for XRS experiments

A class of science problems for which a co-design methodology is critical is that of future detectors employed at DOE facilities that will require at least three orders of magnitude increase in data rates over the next two decades. Bia will focus on two vertical design problems: (1) realistic designs for high-energy physics (HEP) detectors, neuromorphic algorithms for data classification, emerging materials for neuromorphic devices and optical interconnects; and (2) the steering of instruments such as end-stations for X-ray science (XRS) experiments, accelerators for data streaming, emerging materials data streaming accelerators and optical interconnects. 

For each vertical design problem, Bia will utilize AI/ML to develop an extensive, scalable simulation framework to capture the pairwise relationships between the different design levels from an energy efficiency perspective. The resultant design solutions related to new materials, devices and algorithms will advance the science related to HEP and XRS.

By combining advanced materials, innovative devices and specialized computer systems under one framework, Bia aims to reduce the energy used for scientific discovery while also boosting data-processing power.

Bia is a project within the Microelectronics Energy Efficiency Research Center for Advanced Technologies (MEERCAT), a DOE Microelectronics Science Research Center (MSRC) focused on advancing energy-efficient microelectronics through end-to-end co-design.