Computational Model for Chemical Computing
Leveraging chemical reactions for computing has emerged as a promising paradigm. For example, chemical computing using the oscillating Belousov-Zhabotinsky (BZ) reaction can be used to construct simple linearly-bounded Turing machines by storing information in its oscillation state . Due to the low energy consumption and molecular dimension of chemical reactions, the chemical computing paradigm offers the benefit of high energy efficiency and massive parallelism. Our approach of bringing together silicon microfabrication technology and chemical computing is an unexplored yet straightforward path to investigate the potential of miniaturized chemical computing. In first experiments comprising networks of reservoirs with connecting channels, waves of the oscillating chemistry were observed to exhibit complex behaviour depending on the state of the input to the network.
In biological systems, neurons transmit information by exchanging and producing chemical substances, whose dynamics can be modelled using differential equations. The concept of Spiking Neural Unit (SNU)  enables the realization of the biological neural dynamics within the latest AI frameworks. Built around SNUs, the IBM team has developed the NeuroAI Toolkit, which serves as a collection of different neuron models and allows to simulate their behavior individually or in neural networks. One network architecture, that is also believed to be found in the brain, is a so-called “liquid” , consisting of a number of recurrent neurons. An input layer supplies inputs to individual neurons within the liquid, which triggers waves of activity propagating through it, similar to the observed waves in the BZ reaction. An output layer then extracts information from the liquid. Liquids can be used to solve computational tasks, such as the XOR task, larger static classification problems or even more advanced temporal tasks.
In this project, we aim to create a computational model of chemical computing and investigate its computational capabilities using the NeuroAI Toolkit. The specific tasks of this project could entail:
- Understand the chemistry of the liquid
- Understand/Create the differential equations that can be used to model the behavior of the chemistry
- Implement an abstract model as part of the NeuroAI Toolkit
- Verify and compare the simulation with the physical system
- Simulate simple tasks, such as the XOR task or flower classification task (depending on the project duration)
- Perform the same tasks as experiments with the physical system (depending on the project duration)
- Basic experience with neural networks and network architectures
- Experience with the PyTorch framework
- Physico-chemical background (preferred)
IBM is committed to diversity at the workplace. With us you will find an open, multicultural environment. Excellent flexible working arrangements enable all genders to strike the desired balance between their professional development and their personal lives.
How to apply
If you are excited to be a part of IBM Research's mission to shape the future of technology, please submit your CV and motivation reasons.
 M. Dueñas-Dı́ez and J. Pérez-Mercader, “How chemistry computes: language recognition by non-biochemical chemical automata. From finite automata to turing machines,” IScience, vol. 19, pp. 514–526, 2019.
 Woźniak, Stanisław, et al. "Deep learning incorporating biologically inspired neural dynamics and in-memory computing." Nat. Mach. Intell., vol. 2, June 2020, pp. 325-36, doi:10.1038/s42256-020-0187-0.
 Maass, Wolfgang. "Liquid state machines: motivation, theory, and applications." Computability in context: computation and logic in the real world (2011): 275-296.