IP Library › Granted Patent US 12,725,646
Granted Patent B2
US 12,725,646 · App. 18/731,184 · Granted Sep 1, 2026

Ferroelectric in-memory computing

Inventors: Fabia Farlin Athena (Atlanta, GA); Takashi Ando (Eastchester, NY); Paul Michael Solomon (Westchester, NY); Nanbo Gong (White Plains, NY); Guy M. Cohen (Westchester, NY)
Assignee: International Business Machines Corporation
G11C11/2273G11C11/2275G11C11/54
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,725,646
App. No.
18/731,184
Granted
Sep 1, 2026
Kind
B2
Abstract

A computer program product for training a machine learning model. A processor executes program instructions stored on a computer readable media to perform a forward pass read by applying a bias input voltage to a gate of a first resistive processing unit (RPU) configured for infrequent writes and storing a first conductance value representing a synaptic weight value to read from the first RPU a product of the bias input voltage and the first conductance value. The processor performs a backward pass to compute a loss value for the forward pass read and to compute a gradient value to minimize the loss value. The processor applies a coercive input voltage to a gate of a second RPU configured differently for frequent writes to write to the second RPU a second conductance value related to the gradient value and representing a gradient accumulation value.

Claims (38)

1 . A resistive processing unit (RPU) system configured to perform in-memory computing operations for training a machine learning model, the RPU system comprising:

a crossbar array of row control lines intersecting column control lines; and

a plurality of non-volatile resistive memory elements, each connected to one of the row control lines and to one of the column control lines, each non-volatile resistive memory element comprising:

a first ferroelectric field-effect-transistor (FeFET) configured to store a first conductance value that represents a synaptic weight value for the machine learning model; and

a second FeFET configured to store a second conductance value that represents a gradient accumulation value for the machine learning model.

2 . The RPU system of claim 1 , wherein:

the first FeFET comprises:

a first semiconductor substrate, comprising:

a first source region;

a first drain region; and

a first channel region between the first source and first drain regions;

a first ferroelectric layer; and

an interfacial layer between the first channel region and the first ferroelectric layer; and

the second FeFET comprises:

a second semiconductor substrate, comprising:

a second source region;

a second drain region; and

a second channel region between the second source and second drain regions; and

a second ferroelectric layer,

wherein there is no interfacial layer between the second channel region and the second ferroelectric layer.

3 . The RPU system of claim 2 , wherein the first FeFET and the second FeFET are vertically integrated on the first semiconductor substrate and the second semiconductor substrate.

4 . The RPU system of claim 2 , further comprising a first connecting circuit electrically connecting the first source region to the second source region.

5 . The RPU system of claim 4 , further comprising:

a first gate electrode layer on the first ferroelectric layer;

a second gate electrode layer on the second ferroelectric layer, wherein:

at least one of the first source region and the second source region is configured to be connected to a source line;

the first gate electrode layer is configured to be connected to a first gate line;

the second gate electrode layer is configured to be connected to a second gate line;

the first drain region is configured to be connected to a first drain line; and

the second drain region is configured to be connected to a second drain line.

6 . The RPU system of claim 2 , further comprising a second connecting circuit electrically connecting the first drain region to the second drain region.

7 . The RPU system of claim 6 , further comprising:

a first gate electrode layer on the first ferroelectric layer; and

a second gate electrode layer on the second ferroelectric layer, wherein:

at least one of the first source region and the second source region is configured to be connected to a source line;

the first gate electrode layer is configured to be connected to a first gate line;

the second gate electrode layer is configured to be connected to a second gate line; and

at least one of the first drain region and the second drain region is configured to be connected to a drain line.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Jun 1, 2024
From: ATHENA, FABIA FARLIN; ANDO, TAKASHI; SOLOMON, PAUL MICHAEL; GONG, NANBO; COHEN, GUY M.
To: INTERNATIONAL BUSINESS MACHINES CORPORATION
Reel/Frame 067587/0790 →
Continuity (1)
Related Publication 20250372142A1 · Dec 4, 2025
References Cited (24)
US 9922956B2 · Lan · 2018 [cited by applicant]
US 11139401B2 · Doyle · 2021 [cited by applicant]
US 11616057B2 · Majhi · 2023 [cited by applicant]
US 11886987B2 · Das · 2024 [cited by examiner]
US 20210111179A1 · Shriram · 2021 [cited by applicant]
US 20220180156A1 · Gong · 2022 [cited by examiner]
US 20220328502A1 · Yu · 2022 [cited by applicant]
US 20220391681A1 · Gokmen · 2022 [cited by examiner]
EP 3291290A1 · 2018 [cited by applicant]
Cheng, Y. et al., “Emerging Monolithic 3D Integration: Opportunities and Challenges from the Computer System Perspective”, Research Gate (2022), 16 pgs. [cited by applicant]
Choe, G. et al., “3D AND-Type Ferroelectric Transistors for Compute-in-Memory and the Variability Analysis”, IEEE Electron Device Letters (2022), vol. 43:2, pp. 304-307. [cited by applicant]
Dutta, S. et al., “Monolithic 3D Integration of High Endurance Multi-Bit Ferroelectric FET for Accelerating Compute-In-Memory”, IEEE (2023), 4 pgs. [cited by applicant]
Dutta, S. et al., “Lifelong Learning with Monolithic 3D Ferroelectric Ternary Content-Addressable Memory”, IEEE (2023), 4 pgs. [cited by applicant]
Gokmen, T. “Enabling Training of Neural Networks on Noisy Hardware”, Frontiers in Artificial Intelligence (2021), 14 pgs. [cited by applicant]
Gong, N. et al., “Why is Fe—HfO2 more suitable than PZT or SBT for Scaled Nonvolatile 1-T Memory Cell?—A Retention Perspective”, IEEE Electronic Devices Letters (2016), 4 pgs. [cited by applicant]
Hsueh, F. et al., “First Demonstration of Ultrafast Laser Annealed Monolithic 3D Gate-All-Around CMOS Logic and FeFET Memory with Near-Memory-Computing Macro”, IEEE (2023), 4 pgs. [cited by applicant]
Lue, H. et al., “3D AND: a 3D Stackable Flash Memory Architecture to Realize High-Density and Fast-Read 3D NOR Flash and Storage-Class Memory”, IEEE (2023), 4 pgs. [cited by applicant]
Lue, H. et al., “Introduction of 3D AND-type Flash Memory and It's Applications to Computing-in-Memory (CIM)”, IEEE (2023), 2 pgs. [cited by applicant]
Luo, Y. et al., “Monolithic 3D Compute-in-Memory Accelerator with BEOL Transistor based Reconfigurable Interconnect”, IEEE (2021), 4 pgs. [cited by applicant]
Ni, K. et al., “Critical Role of Interlayer in Hf0.5Zr0.5O2 Ferroelectric FET Nonvolatile Memory Performance”, IEEE Transactions on Electron Devices (2018), vol. 65:6, pp. 2461-2469. [cited by applicant]
Prakash, O. et al., Monolithic 3D Integrated BEOL Dual-Port Ferroelectric FET to Break the Tradeoff Between the Memory Window and the Ferroelectric Thickness, IEEE (2023), 4 pgs. [cited by applicant]
Wang, P. et al., “Drain-Erase Scheme in Ferroelectric Field Effect Transistor—Part I: Device Characterization”, IEEE Transactions On Electron Devices (2020), 7 pgs. [cited by applicant]
Ye, H. “High-Performance Back-End-Of-Line Compatible Indium Tungsten Oxide (Iwo) Transistors For Monolithic-3d Application”, A Dissertation, submitted to the University of Notre Dame (2022), 24 pgs. [cited by applicant]
Yin, X. et al., “FeCAM: A Universal Compact Digital and Analog Content Addressable Memory Using Ferroelectric”, IEEE Transactions On Electron Devices (2020), vol. 67:7, pp. 2785-2792. [cited by applicant]