IP Library › Granted Patent US 12,554,508
Granted Patent B2
US 12,554,508 · App. 17/937,101 · Granted Feb 17, 2026

Processing complex packed tensors using integrated circuit of real and complex packed tensors in complex domain

Inventors: Hayim Shaul (Kfar Saba, IL); Nir Drucker (Zichron Yaakov, IL); Ehud Aharoni (Kfar Saba, IL); Omri Soceanu (Haifa, IL); Gilad Ezov (Nesher, IL)
Assignee: International Business Machines Corporation
G06F9/3887G06F9/30036H04L9/008
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 12,554,508
App. No.
17/937,101
Granted
Feb 17, 2026
Kind
B2
Abstract

An example system includes a processor that can receive a number of complex packed tensors, wherein each of the complex packed tensors include real numbers encoded as imaginary parts of complex numbers. The processor can execute a single instruction, multiple data (SIMD) operation on the complex packed tensors using an integrated circuit of real and complex packed tensors in a complex domain to generate a result.

Claims (32)

1 . A computer system comprising:

a processor set;

one or more computer-readable storage media; and

program instructions stored on the one or more computer-readable storage media to cause the processor set to perform operations comprising:

receiving a plurality of non-complex packed tensors and complex packed tensors, wherein each of the complex packed tensors comprises real numbers encoded as imaginary parts of complex numbers;

transforming a non-complex packed tensor into a complex-packed tensor of the plurality of complex packed tensors; and

executing a homomorphically encrypted single instruction, multiple data (SIMD) operation on the complex packed tensors using an integrated circuit of real and complex packed tensors corresponding to SIMD-based registers in a complex domain to generate a result.

2 . The system of claim 1 , wherein the complex packed tensors comprise complex packed tile tensors.

3 . The system of claim 1 , wherein the homomorphically encrypted SIMD operation is executed securely using a fully homomorphic encryption scheme and the result comprises an encrypted result.

4 . The system of claim 1 , wherein the complex packed tensors correspond to ciphertexts.

5 . The system of claim 1 , wherein the complex packed tensors correspond to the SIMD-based registers of a hardware processing unit.

6 . The system of claim 1 , wherein the plurality of complex packed tensors correspond to a single data set packed using imaginary parts of complex numbers to represent half of the single data set.

7 . The system of claim 1 , wherein the plurality of complex packed tensors correspond to two data sets with similar dimensions, and a first data set is represented by real parts of complex numbers and a second data set is represented by imaginary parts of the complex numbers.

8 . The system of claim 1 , wherein the SIMD operation approximates a function of a layer of a neural network.

9 . The system of claim 8 , wherein the layer of the neural network comprises a convolutional layer.

10 . The system of claim 8 , wherein the layer of the neural network comprises a fully-connected layer.

11 . The system of claim 1 , wherein the non-complex packed tensor is transformed using a complex packing method.

12 . A computer-implemented method, comprising:

receiving, via a processor, a plurality of non-complex packed tensors and complex packed tensors, wherein each of the complex packed tensors comprise real numbers encoded as imaginary parts of complex numbers;

transforming a non-complex packed tensor into a complex-packed tensor of the plurality of complex packed tensors; and

executing, via the processor, a homomorphically encrypted single instruction, multiple data (SIMD) operation on the complex packed tensors using an integrated circuit of real and complex packed tensors corresponding to SIMD-based registers in a complex domain to generate a result.

13 . The computer-implemented method of claim 12 , wherein executing the SIMD operation comprises performing a dot product operation on the plurality of complex packed tensors.

14 . The computer-implemented method of claim 12 , wherein executing the SIMD operation comprises performing a matrix-vector multiplication using an operation between a ciphertext corresponding to a vector and a set of values of an encrypted matrix corresponding to an underlying plaintext matrix, wherein each row of the encrypted matrix is complex packed in an additional ciphertext.

15 . The computer-implemented method of claim 12 , comprising adding complex packed dimensions to a tile tensor shape information, and additional operators a set of operators supported by a tile tensor, wherein executing the SIMD operation comprises performing an operation on a set of complex packed tile tensors.

16 . The computer-implemented method of claim 12 , comprising generating the plurality of complex packed tensors using an interleaving complex packing operation on pairs of ciphertexts, and in response to detecting an odd number of ciphertexts, leaving a last ciphertext non-complex-packed.

17 . The computer-implemented method of claim 12 , wherein executing the SIMD operation comprises performing a dot product operation, wherein the plurality of complex packed tensors comprise a first vector encoded with odd indexed elements as real parts and a negation of event indexed elements as imaginary parts and a second vector encoded with odd indexed elements as real parts and even indexed elements as imaginary parts, wherein performing the dot product operation comprises computing a dot product of the first vector and the second vector and outputting a real part of the dot product.

18 . The computer-implemented method of claim 12 , wherein executing the SIMD operation comprises performing a diagonalization operation comprising a sum operation implemented on a complex packed vector with a non-complex-packed matrix using a rotate-one operator.

19 . A computer program product for packing real numbers, the computer program product comprising a computer-readable storage medium having program code embodied therewith, the program code executable by a processor to cause the processor to:

receive a plurality of non-complex packed tensors and complex packed tensors, wherein each of the complex packed tensors comprise real numbers encoded as imaginary parts of complex numbers;

transform a non-complex packed tensor into a complex-packed tensor of the plurality of complex packed tensors; and

execute a homomorphically encrypted single instruction, multiple data (SIMD) operation on encrypted complex packed tensors using an integrated circuit of real and complex packed tensors corresponding to SIMD-based registers in a complex domain to generate a result.

20 . The computer program product of claim 19 , wherein the complex packed tensors and the generated result are encrypted, further comprising program code executable by the processor to send the result to a client device to be decrypted.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Oct 3, 2022
From: SHAUL, HAYIM; DRUCKER, NIR; AHARONI, EHUD; SOCEANU, OMRI; EZOV, GILAD
To: INTERNATIONAL BUSINESS MACHINES CORPORATION
Reel/Frame 061286/0560 →
Continuity (1)
Related Publication 20240126557A1 · Apr 18, 2024
References Cited (20)
US 20160164671A1 · Gentry · 2016 [cited by examiner]
US 20170134157A1 · Laine et al. · 2017 [cited by applicant]
US 20190007196A1 · Malluhi et al. · 2019 [cited by applicant]
US 20190250915A1 · Yadavalli · 2019 [cited by examiner]
US 20190385063A1 · Yu · 2019 [cited by examiner]
US 20200076570A1 · Musuvathi et al. · 2020 [cited by applicant]
US 20200105046A1 · Wald · 2020 [cited by examiner]
US 20210042127A1 · Koneru · 2021 [cited by examiner]
US 20210042624A1 · Matveev · 2021 [cited by examiner]
US 20210182021A1 · Wang · 2021 [cited by examiner]
US 20210191765A1 · Bokam · 2021 [cited by examiner]
US 20210357475A1 · Wang · 2021 [cited by examiner]
US 20220126210A1 · Kumar · 2022 [cited by examiner]
US 20220138286A1 · Zage · 2022 [cited by examiner]
US 20220197601A1 · Adelman · 2022 [cited by examiner]
US 20220413848A1 · Pal · 2022 [cited by examiner]
Chen et al., “Double Batch for RLWE-Based Leveled Fully Homomorphic Encryption”, Jul. 2015, vol. 24, No. 3, pp. 661-666 (Year: 2015). [cited by examiner]
Aharoni et al., “Tile Tensors: A versatile data structure with descriptive shapes for homomorphic encryption”, arXiv:2011.01805v1 [cs.CR], Nov. 3, 2020, 17 pages. [cited by applicant]
Duhyeong Kim et al., “Approximate Homomorphic Encryption over the Conjugate-invariant Ring”, In: Lee K. (eds) Information Security and Cryptology—ICISC 2018. ICISC 2018. Lecture Notes in Computer Science, vol. 11396, 18… [cited by applicant]
Fabian Boemer et al., “nGraph-HE2: A High-Throughput Framework for Neural Network Inference on Encrypted Data”, In Proceedings of the 7th ACM Workshop on Encrypted Computing & Applied Homomorphic Cryptography (WAHC'19).… [cited by applicant]