PNNL

Before digital computers existed, analog devices were used to assist humans in making calculations, predictions, and measurements. From the ancient Antikythera mechanism used to predict movements of celestial bodies in 200 B.C. to the ‘Brass Brains’ Tide Predicting Machine No. 2 used by the U.S. Coast and Geodetic Survey from 1910 to 1965, many devices aided in computational tasks before being replaced by digital computers.

Though it is exceedingly rare for an ousted technology to reemerge, analog computing appears to be doing just that. With transistor sizes quickly approaching the limit of Moore’s Law—meaning engineers will no longer be able to increase computer speeds by decreasing transistor sizes—researchers need to find new ways to increase energy efficiency in computing. 

Antonino Tumeo, who is a computer scientist at Pacific Northwest National Laboratory (PNNL), recently co-led the 2024 Workshop on Basic Research Needs for Analog Computing for Science to explore how analog computing may offer such efficiencies. This workshop was supported by the Department of Energy, Advanced Scientific Computing Research program.

“Analog computers hold the potential to transform computing yet again,” said Tumeo. “With analog computing’s increased specificity and energy efficiency, scientists may surpass the limitations of Moore’s Law to meet future computing needs.”

During the workshop, participants explored different aspects of analog computing, including its mathematical foundation, computational memory, and co-design of analog and digital machines. The workshop also delved into analog computing using chemical and biological molecules. At PNNL, Tumeo leads ChemComp, an exploratory research project that uses chemical reaction networks to solve differential equations. 

“Chemical reactions themselves constitute an instruction set, just like those for the typical operations of a computer,” said Tumeo. “And cells naturally optimize reactions using the lowest energy possible, thus are capable of solving certain optimization problems very efficiently.” 

As noted in the pre-workshop read ahead document, a potential application of biochemical computing includes the development of “smart drugs” that can detect disease and autonomously respond with therapeutics. Other areas where analog computing can contribute include analog integrated circuits for scientific instrumentation in high-energy physics and material science and probabilistic computing for complex problems, such as climate models and neuromorphic systems. 

The next-generation of analog computing holds great promise to meet future energy-efficient computing needs.