NSF-SNSF: Chemical Computing Architectures (CheCoA)

Current Artificial Intelligence (AI) implementations in semiconductor-based computing devices are energy-hungry, consuming resources at an enormous scale. Developing computing schemes with a significantly smaller energy consumption and ecological footprint than used in the current systems has immense societal benefits. In this work, first prototypes of miniaturized chemical computing devices will be demonstrated and a pathway towards neural networks that can be trained by large data sets with a fraction of today's energy consumption will be established.

The group will also train a new generation of scientists seeking to unlock the power of biologically inspired computation.

Developing computing schemes with a significantly smaller energy consumption and ecological footprint than used in the current systems has immense societal benefits. Chemical computing devices may provide low energy computing devices. Many examples of chemical computing devices are based on macroscopic reaction volumes of the order of 10^23 molecules/mL.

An energy-efficient architecture will operate at an optimal number of molecules that produce signals with acceptable levels of noise. Today, it is not clear what minimal volumes of computing chemical reaction networks (CRNs) and their reactors can be achieved and whether fast and sensitive read-out and input methods for ultra-low amount of chemicals are available.

Furthermore, effective coupling schemes need to be adapted to the microscale (inhibition and acceleration of reaction paths). Innovative ways of dynamically changing a miniaturized chemical computing architecture – an important prerequisite for complex computations – are not existent at this time and need to be discovered and developed. This award will allow investigators at Harvard, together with investigators at IBM Research in Zurich, to produce miniaturized reactor arrays and microfluidic supply structures for coupled CRN nodes of picoliter and femtoliter volumes (10^14 and 10^11 molecules).

Their fabrication will rely on IBM's well-established microfabrication capabilities in silicon-based lithography, but also on unconventional methods such as thermal scanning probe lithography (t-SPL) (developed by the IBM team). The implementation of coupling schemes of CRNs in different reactors for the purpose of complex computation will rely on the principal investigator's renowned expertise in the chemistry of oscillating CRNs.

The group will contribute physicochemical state variables and condensed matter physics description of the system that are required for the correlation between input and output of the computing CRNs. The team ideally combines strong experience and expertise in the relevant fields of this work (physics, chemistry, microfabrication and computing theory).

The team at Harvard will be supported by this NSF award. The team at IBM Research will be supported by the Swiss National Science Foundation (SNSF).

This collaborative U.S.-Swiss project is supported by the U.S. National Science Foundation (NSF) and the Swiss National Science Foundation (SNSF), where NSF funds the U.S. investigator and SNSF funds the partners in Switzerland.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.