The field of chemical computing has a huge potential but still many unknowns. By their nature, molecules can perform complex tasks such as molecular recognition and chemical reactions at the smallest possible footprint and energy scales. Chemical reactions can further be cascaded and they are typically highly non-linear. If we can harvest these fascinating features and direct them towards more general information processing, and finally computing, there might be a huge promise for high-performance and low-power machines.
Oscillatory Chemical Reactions
The first implementation is based on oscillatory chemical reactions for which a variety of early computing implementations exist. A single chemical volume of such chemistry was recently shown to implement automata that recognize simple formal languages. This result clearly demonstrates that chemistry can perform complex tasks and store multiple states of information. Here, we ask the question of whether we can harness this functionality by using many miniaturized chemical volumes to create highly connected chemical information entities that solve programmed problems in a highly parallel and energy efficient manner.
Such an implementation will serve to answer several important and fundamental questions about the role of molecules for information processing:
- Is oscillatory chemistry sufficiently robust and reproducible?
- To how many neighboring volumes can a chemical volume connect?
- What are the fundamental scaling limits of creating miniaturized chemical volumes in terms of size and rates of oscillations?
- Can we tailor the necessary inhibitory and excitatory couplings between neighboring volumes so that a programming task can be performed in a flexible manner?
In our implementation of a chemical Ising solver. we will concentrate on the Belousov-Zhabotinsky reaction. This is an oscillatory chemical reaction that shows rich oscillating behavior, which can be manipulated by a variety of chemical and physical inputs. We will prepare microscale reaction compartments using lithographic methods and study the interactions between the compartments, mediated by diffusive coupling within the matrix (inhibitory) and through the reaction mixture (excitatory).
Complex Chemical Reaction Networks (CRNs)
The second implementation is based on complex chemical reaction networks (CRNs). By mixing a few feedstock molecules and reagents together, these networks develop on their own and thereby produce a large amount of chemically diverse product outputs. One example is the well-known formose reaction for which one of our partners within CoreNet has recently demonstrated that the CRN responds susceptibly and non-linearly to variations in its chemical input composition (see the reference here). Together with a stringent control of the conditions to perform the chemical reactions inside reactors, we aim to link all the inputs of the wet-chemistry experiment to digital inputs to transmit the computing problem. One big challenge is the timely read-out of the CRN state that is used as the system’s response. With dedicated algorithms to extract the essential features from on-line and off-line analytics, we can finally operate a chemical computer without human intervention 24/7 and tackle first use cases.
Within this project, we aim to answer the following fundamental questions:
- Can automation govern sufficient control over the outputs of CRNs?
- Do CRNs behave chaotically or do their reaction pathways follow certain rules?
- Is the self-organization and self-limitation within CRNs a feature to be used for reproducible chemical computing?
- Can we use outputs from one CRN to trigger another CRN to develop?
- How fast can we encode the system and what is the typical latency?
With our group’s extensive experience in micro- and nano-fabrication, nano-fluidics, surface chemistry, flow-chemistry, and analytics - combined with our solid background in chemistry and physics, we are confident that we will be able to successfully implement the required components and demonstrate chemical computing using the hard- and software described above. Outcomes will be a miniaturized, purely chemical, Ising solver that can tackle NP hard problems such as the traveling salesperson problem or voice and image recognition.
Projects & funding
This work is funded by the European Union and supported by the Swiss State Secretariat for Education, Research and Innovation (SERI) under contract numbers 22.00017 and 22.00034.
Within CoreNet, we collaborate with Radboud University Nijmegen (Prof. Wilhelm Huck), Universidad Autónoma de Madrid (Prof. Andrès de la Escosura), University of Southern Denmark (Prof. Daniel Merkle) and Consejo Superior de Investigaciones Cientificas Spain (Prof. Alejandro Cifuentes).