In-memory mechanical computing

Area of Science:

• Mechanical Engineering
• Computer Science
• Materials Science

Background:

• Mechanical computing relies on adaptable materials for sensing, actuation, and deformation control.
• Existing mechanical computing modules face limitations due to inefficient memory access and signal propagation.

Purpose of the Study:

• To develop a novel in-memory mechanical computing architecture.
• To overcome limitations in data traffic and signal propagation in mechanical computing.
• To enable function-complete and neuromorphic computing within mechanical memory units.

Main Methods:

• Developed an in-memory mechanical computing architecture.
• Integrated computing within the interaction network of mechanical memory units.
• Utilized 3D printing for fabricating mechanical computers.

Main Results:

• Demonstrated function-complete and neuromorphic computing.
• Reduced data traffic and simplified data exchange.
• Experimentally realized a reprogrammable mechanical binary neural network, a mechanical self-learning perceptron, and all 16 possible two-input logic gates.

Conclusions:

• The in-memory mechanical computing architecture facilitates efficient data processing and memory access.
• This architecture enables the design and fabrication of intelligent mechanical systems.
• The demonstrated capabilities pave the way for advanced mechanical computing applications.