A stochastic encoder using point defects in two-dimensional materials

Area of Science:

• Materials Science
• Quantum Computing
• Neuromorphic Engineering

Background:

• Electronic device defects are typically detrimental but offer potential in quantum and energy applications.
• The use of defects for novel computational paradigms like neuromorphic computing is largely unexplored.

Purpose of the Study:

• To leverage defects in scaled 2D semiconductor transistors for a stochastic inference engine.
• To investigate the role of point defects in tungsten diselenide (WSe2) field-effect transistors (FETs) and their impact on random telegraph noise (RTN).

Main Methods:

• Atomistic imaging
• Density functional theory (DFT) calculations
• Device modeling
• Low-temperature transport experiments
• Spiking neural network (SNN) implementation

Main Results:

• Detailed characterization of point defects in WSe2 FETs and their influence on RTN.
• Construction of a stochastic encoder using RTN.
• Demonstrated enhanced inference accuracy on noisy medical-MNIST images compared to deterministic encoders.

Conclusions:

• Intrinsic point defects in 2D materials can be effectively utilized as a resource for neuromorphic computing.
• Stochastic inference engines based on RTN show promise for noise-resilient computation.