As the energy requirement of large machine learning models continues to increase, new brain-like (neuromorphic) computing architectures that reduce power consumption are highly sought after. When looking to the human brain as the ultimate inspiration for low power computation, we can emulate various functionalities of a neuron.
Organic Synapse: The Electrochemical Random Access Memory
In a traditional Von-Neumann architecture, most of the energy cost of computation comes from shuttling large amounts of information from the CPU to the memory. Thus, combining memory and computation into a single device can significantly reduce energy costs. Traditional semiconductor memory technology has yet to satisfy the needs of the “artificial synapse” that is the core of the neuromorphic computing architecture. In our group, we use organic semiconductors to mimic synaptic behavior in an electrochemical organic neuromorphic device, which couples ionic and electronic currents to emulate the strength of neuron-to-neuron connections. The high linearity and low switching energy of the memories make them highly suitable for massively parallel neural algorithm accelerators, i.e. brain-like computer chips. Our group’s research focuses on leveraging the ionic/electronic transport properties of polymeric semiconductors to design novel devices for neuromorphic computing.
Organic Spiking Neuron
Spiking neurons mimic the communication patterns of biological neurons, enabling efficient information processing by encoding data in temporal spikes, which offers higher information density compared to traditional encoding schemes. This temporal encoding facilitates event-driven processing and lower power consumption. All organic Axon Hillock circuits are being developed in the group. We are working to implement spike-frequency adaptation on organic hardware by exploiting the short-term plasticity property of OECTs and to build an organic artificial neuron with tunable spike-frequency adaptation. Additionally, these circuits are being integrated with novel sensors.
As of 2024, work on this project is relatively new and ongoing. Check back later for updates on our progress!
Sequence Decoding Dendrite
Devices adept at recognizing sequences are pivotal for the realization of scalable 3D architectures. Our group is actively developing a dendrite device that can decode sequences pulsed at the gates of a synapse-like device. In order to fabricate a dendrite, new fabrication methods were developed in order to micropattern a solid-state electrolyte on top of micropatterned conducting polymer materials.
As of 2024, work on this project is new and ongoing. Check back later for updates on our progress!
Active members in this area: Kalee Rozylowicz, Yeongmin Park, Julian Mele, Gerwin Dijk