Ferroelectric Diode Arrays for Reconfigurable In-Memory Computing

Field-programmable ferroelectric diode arrays enable nonvolatile storage, search, and neural-network operations using transistor-free memory and synapse architectures.
Tech Title: Technology:
Technology: The system includes arrays of ferroelectric diodes formed atop semiconductor structures such as CMOS wafers. Memory arrays can use a V/2 scheme for readout and programming. TCAM cells can use two oppositely polarized diodes in 0T-2R structures. Neural-network implementations use multi-bit diode synapses with multiple conductive states programmed by electrical pulses.
Problem Title: Problem:
Problem: Data-intensive computing can require separate architectures for storage, search, and matrix multiplication. Integrating these functions on one chip has been challenging. Conventional in-memory computing designs can add device overhead and complicate compact implementation.
Solution Title: Solution:
Solution: The technology uses field-programmable ferroelectric diodes built from materials such as doped aluminum nitride or hafnium zirconium oxide. These diodes are non-volatile and can be pulsed to pulse-number-dependent analog states. The architecture supports transistor-free memory cells, TCAM cells, and multi-bit diode synapses for reconfigurable in-memory computing.
Advantages:

Supports nonvolatile operation with field programmability through electrical pulsing.
Enables transistor-free cell architecture for storage and search.
Can be implemented using ferroelectric diode materials including doped aluminum nitride or hafnium zirconium oxide.
Supports multiple conductive states for synaptic operation in neural-network arrays.
Compatible with CMOS back-end-of-line (BEOL) processing.

Applications:

Nonvolatile Memory: Provides field-programmable memory cells using ferroelectric diode arrays.
Search Hardware: Implements TCAM cells using two oppositely polarized diodes in 0T-2R structures.
Neural Networks: Uses multi-bit diode synapses for array-based neural-network computation.
Reconfigurable Computing: Supports storage, search, and neural-network functions within reconfigurable in-memory-computing architectures.

Intellectual Property: