Neuromorphic

Contact: Alexandre ZIMMERS

The brain-inspired codes of the AI revolution primarily run on conventional silicon-based computer architectures that were not designed for this purpose. This will soon lead to unsustainable energy consumption as AI continues to grow. Neuromorphic architectures offer the promise of lower energy consumption by mimicking the brain’s basic components—neurons and synapses—ideally using a single material. Unfortunately, among the few quantum materials that naturally act as spiking “neuristors” (artificial neurons), non-volatile “synaptor” memory (artificial synapses) has been hard to implement (see Fig. 1).

Fig. 1: Schematic illustration of a neuromorphic system. (a) Biological model: the neuron soma receives inputs via synapses. (b) Bio-inspired electronic model: an electronic neuristor accumulates inputs generated by multiple pre-synaptic neuristors, with weights modulated by memristive synaptors. Bottom: VO₂ naturally functions as an artificial neuron, but its use as an artificial synapse remains to be established.

Vanadium dioxide (VO₂) is known to be a good neuristor material operating around room temperature. To scale a neuromorphic cuircuit on a single chip, one also needs to turn VO₂ into an artificial synapse. To do so, we first studied its electronic patterns during the insulator–metal transition. This transition is particularly interesting, as it exhibits fractal electronic domains spanning several orders of magnitude. To do so we have developed a new optical microscopy method that allows precise sub-micron recording of these patterns (see Fig. 2). This series of clustered images enabled us to generate Tc maps [1] and to reveal the underlying interactions using newly developed machine-learning techniques [2].

Fig. 2: Multiscale electronic patterns in VO₂ during the insulator–metal transition [1].

Using this knowledge, over the years we have created synaptic behavior in VO₂ in various ways: by utilizing temperature sweeps [3], a focused laser beam and an AFM scanning tip [4], and Au nanodisks on VO₂ [5].

Fig. 3: "Ramp Reversal Memory" in VO₂ [3].

Fig. 4: Local control of VO₂ insulator–metal patterns using temperature sweeps, a focused laser beam, an AFM tip [4].

Fig. 5: The insulator–metal transition in a VO₂ thin film is facilitated by ordered arrays of gold (Au) nanodisks [5].

These advances offer, for the first time, the possibility of creating rewritable synaptic connections between neuristors in a single-material VO₂ neuromorphic chip.

References:

[1] Optical Mapping and On-Demand Selection of Local Hysteresis Properties in VO₂
M. Alzate Banguero, S. Basak, N. Raymond, F. Simmons, P. Salev, I. K. Schuller, L. Aigouy, E. W. Carlson, A. Zimmers. Condensed Matter 10 (1), 12 (2025)

[2] Deep learning Hamiltonians from disordered image data in quantum materials
S. Basak, M. A. Banguero, L. Burzawa, F. Simmons, P. Salev, L. Aigouy, M. M. Qazilbash, I. K. Schuller, D. N. Basov, A. Zimmers, E. W. Carlson. Physical Review B 107, 205121 (2023)

[3] Spatially Distributed Ramp Reversal Memory in VO₂
S. Basak, Y. Sun, M. A. Banguero, P. Salev, I. K. Schuller, L. Aigouy, E. W. Carlson, A. Zimmers. Advanced Electronic Materials 9 (10), 2300085 (2023)

[4] Tuning the Resistance of a VO₂ Junction by Focused Laser Beam and Atomic Force Microscopy
Z. Fang, M. Alzate-Banguero, A. R. Rajapurohita, F. Simmons, E. W. Carlson, Z. Chen, L. Aigouy, A. Zimmers. Advanced Electronic Materials 11 (2), 2400249 (2025)

[5] Plasmon-enhanced photothermal sensing through coupled VO₂/Au nanodisks
Z. Fang, A. Zimmers, Z. Chen, L. Billot, A. García-Martín, L. Aigouy. Surfaces and Interfaces 62, 106145 (2025)