The 3D Monolithically Integrated Hardware Based Neural System with Enhanced Memory Window of the Volatile and...

Area of Science:

• Materials Science
• Neuroscience
• Electrical Engineering

Background:

• 3D neuromorphic hardware systems are crucial for advanced AI applications.
• Wafer bonding and 3D integration can degrade synaptic device memory windows due to process defects and thermal stress.

Purpose of the Study:

• To address memory window degradation in 3D integrated neuromorphic devices.
• To improve synaptic device performance for hardware-based neural systems.

Main Methods:

• Investigated Ag diffusion in Ta2O5 and HfO2 for volatile memristors.
• Studied interconnection and gate metal (Ru) to reduce interface traps in non-volatile memory devices.
• Integrated 12x14 array devices with CMOS circuits using wafer bonding.

Main Results:

• Achieved a memory window improvement exceeding 10^6 in both volatile and non-volatile devices.
• Demonstrated recognizable image recognition of letters with 92% accuracy using trained array devices.
• Validated synaptic characteristics and weight changes in 3D integrated array devices.

Conclusions:

• Successfully modulated memory windows up to 10^6 in an integrated 3D neuromorphic hardware system.
• The proposed methods mitigate device degradation, enabling robust neuromorphic applications.
• The study highlights the potential of 3D integration for advanced, high-performance neural systems.