IP Library Granted Patent US 12,430,397
Granted Patent B2
US 12,430,397 · App. 17/451,550 · Granted Sep 30, 2025

Convolutions with optical finite impulse response filters

Inventors: Pascal Stark (Thalwil, CH); Folkert Horst (Wettingen, CH); Roger F. Dangel (Zug, CH); Bert Jan Offrein (Schoenenberg, CH); Lorenz K. Muller (Dubendorf, CH)
Assignee: International Business Machines Corporation
G06F17/153H03H2017/0081H03H2218/12
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 12,430,397
App. No.
17/451,550
Granted
Sep 30, 2025
Kind
B2
Abstract

A method of processing data and related apparatuses. The method relies on an optical finite impulse response (FIR) filter. This optical FIR filter comprises several delay stages having weights set in accordance with parameters of a transformation to be applied by the optical FIR filter. Each of the delay stages is configured to impose a delay matched to a given input data period corresponding to a given input sample rate. According to the method, an optical signal is coupled into the optical FIR filter. The optical signal carries a data stream of input samples encoded at the given input sample rate; the data stream represents the data to be processed. Next, output samples are collected from an output data stream carried by an output optical signal obtained in output of the optical FIR filter. A set of output samples are obtained, which are representative of processed data.

Claims (57)

1. A method of processing data, wherein the method comprises:

providing an optical finite impulse response (FIR) filter, wherein the optical FIR filter comprises several delay stages having weights set in accordance with parameters of a transformation to be applied by the optical FIR filter, the delay stages configured to impose a delay matched to a given input data period corresponding to a given input sample rate, and wherein the optical FIR filter is a lattice filter in two-port lattice form, the delay stages being four-terminal lattice filter stages that are optically connected in series, and the four-terminal lattice filter stages comprising two waveguides and two phase shifters, the two phase shifters formed at respective portions of the two waveguides;

coupling an optical signal into the optical FIR filter, wherein the optical signal carries a data stream of input samples encoded at the given input sample rate, the data stream representing data to be processed; and

collecting output samples in an output data stream carried by an output optical signal obtained in output of the optical FIR filter to obtain a set of output samples that are representative of processed data.

2. The method according to claim 1 , wherein the method further comprises;

reconstructing transformed data based on the set of the output samples obtained.

3. The method according to claim 1 , wherein

the output samples are collected by sub-sampling the output samples in the output data stream, at a sub-sampling rate determined in accordance with the parameters of the transformation applied by the optical FIR filter.

4. The method according to claim 1 , wherein

the transformation to be applied is an r-dimensional convolution, r≥1, having a kernel of size k r with given kernel values.

5. The method according to claim 4 , wherein

r≥2,

the optical FIR filter comprises at least k r −1 delay stages, the weights of which are set in accordance with the given kernel values, and

the output samples are collected by sub-sampling the output samples in the output data stream at a sub-sampling rate of k 1-r , whereby the set of output samples obtained is a selection of the output samples carried in the output data stream, the selection including every k r-1 -th sample of at least a subset of the output samples in the output data stream.

6. The method according to claim 4 , wherein

data to be transformed are d-dimensional data, d≥2,

the method further comprises, prior to coupling the optical signal into optical input of the optical FIR filter, scanning through the d-dimensional data according to a predetermined scan path, and

the weights of the delay stages are set in accordance with the given kernel values, consistently with the predetermined scan path.

7. The method according to claim 1 , wherein

the method further comprises, prior to coupling the optical signal into optical input of the optical FIR filter, setting the weights in accordance with the parameters of the transformation to be applied.

8. The method according to claim 1 , wherein

the method further comprises, prior to setting the weights, mapping the weights in accordance with the parameters of the transformation to be applied, whereby the weights are set in accordance with the mapping of the weights.

9. The method according to claim 1 , wherein

the four-terminal lattice filter stages comprise optical phase shifters, and

the method further comprises, prior to coupling the optical signal into optical input of the optical FIR filter, tuning the optical phase shifters to set the weights in accordance with the parameters of the transformation to be applied.

10. An apparatus for processing data, wherein the apparatus comprises:

an optical finite impulse response (FIR) filter comprising several delay stages having weights set in accordance with parameters of a transformation to be applied by the optical FIR filter, the delay stages configured to impose a delay matched to a given input data period corresponding to a given input sample rate, wherein the optical FIR filter is a lattice filter, and the lattice filter is in two-port lattice form, the delay stages being four-terminal lattice filter stages that are optically connected in series, the four-terminal lattice filter stages comprising two waveguides and two phase shifters, the two phase shifters formed at respective portions of the two waveguides;

a transmitter connected to an optical input of the optical FIR filter and configured to couple an optical signal into the optical FIR filter, wherein, in operation, the optical signal carries a data stream of input samples encoded at the given input sample rate, the data stream representing data to be processed; and

a receiver connected to an optical output of the optical FIR filter and configured to collect output samples in an output data stream carried by an output optical signal obtained in output of the optical FIR filter to obtain a set of the output samples that are representative of the transformation.

11. The apparatus according to claim 10 , wherein

the apparatus further comprises a data reconstruction unit connected to the receiver and configured to reconstruct transformed data based on the set of the output samples obtained.

12. The apparatus according to claim 10 , wherein the receiver is further configured to collect the output samples by sub-sampling the output samples in the output data stream, at a sub-sampling rate determined in accordance with the parameters of the transformation applied by the optical FIR filter.

13. The apparatus according to claim 10 , wherein

in the four-terminal lattice filter stages,

the two waveguides are arranged so as to form two directional couplers, and

a length difference between optical paths enabled by the two waveguides is such that a propagation delay difference between the optical paths corresponds to the given input data period.

14. The apparatus according to claim 13 , wherein, in the four-terminal lattice filter stages,

the two waveguides correspond to a first waveguide and a second waveguide;

the respective portions of the two waveguides correspond to a first portion of the first waveguide and a second portion of the second waveguide, respectively;

a first phase shifter of the two phase shifters is arranged at the first portion of the first waveguide between the two directional couplers;

a second phase shifter of the two phase shifters is arranged at the second portion of the second waveguide, the second portion being in output of the second waveguide; and

the four-terminal lattice filter stages are connected in an alternated fashion, such that the first waveguide and the second waveguide of one of the four-terminal lattice filter stages each connects to the second waveguide and the first waveguide, respectively, of a next one of the four-terminal lattice filter stages.

15. The apparatus according to claim 14 , wherein

the first waveguide is a longest one of the two waveguides, and

the two directional couplers and the first phase shifter are configured as a phase-tunable Mach-Zehnder interferometer.

16. The apparatus according to claim 14 , wherein

the two phase shifters are thermo-optic phase shifters;

the apparatus further comprises an electrical circuit, which includes heaters that form part of the thermo-optic phase shifters;

the heaters are in thermal communication with said respective portions of the two waveguides; and

the electrical circuit is configured to apply an electrical current to the heaters of the delay stages, so as to be able to set the weights of the delay stages.

17. The apparatus according to claim 16 , wherein

the apparatus comprises a silicon photonic chip including the optical FIR filter.

18. The apparatus according to claim 17 , wherein

the silicon photonic chip has a layer structure, which includes:

a silicon-insulator-silicon layer stack, including a silicon handle wafer, an oxide layer on top of the silicon handle wafer, and a topmost silicon layer on top of the oxide layer, the topmost silicon layer structured so as to form the two waveguides of the four-terminal lattice filter stages;

an electrically insulating layer cladding the topmost silicon layer, the heaters arranged on top of the electrically insulating layer, and arranged at the respective portions of the two waveguides; and

electrical leads and electrical pads arranged on top of the electrically insulating layer, wherein the electrical leads connect the heaters to the electrical pads.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Oct 20, 2021
From: STARK, PASCAL; HORST, FOLKERT; DANGEL, ROGER F.; OFFREIN, BERT JAN; MULLER, LORENZ K.
To: INTERNATIONAL BUSINESS MACHINES CORPORATION
Reel/Frame 057850/0413 →
Continuity (1)
Related Publication 20230118621A1 · Apr 20, 2023
References Cited (21)
US 8976445B1 · Willner et al. · 2015 [cited by applicant]
US 20170054510A1 · Zortea et al. · 2017 [cited by applicant]
US 20190019100A1 · Roques-Carmes et al. · 2019 [cited by applicant]
US 20190370644A1 · Kenney et al. · 2019 [cited by applicant]
US 20200363660A1 · Nakajima · 2020 [cited by examiner]
US 20210192325A1 · Hoang · 2021 [cited by examiner]
US 20240250755A1 · Nakajima · 2024 [cited by examiner]
CN 110516802A · 2019 [cited by applicant]
WO 2020102204A1 · 2020 [cited by applicant]
Ahmed et al., “Integrated Photonic FFT for Optical Convolutions towards Efficient and High-Speed Neural Networks,” https://arxiv.org/vc/arxiv/papers/2002/2002.01308v1.pdf, 12 pgs. [cited by applicant]
Bagherian et al., “On-Chip Optical Convolutional Neural Networks,” arXiv:1808.03303v 2 [cs.ET] Aug. 16, 2018, 18 pgs. [cited by applicant]
Farabet, et al., “NeuFlow_A Runtime Reconfigurable Dataflow Processor for Vision,” CVPR 2011 Workshops, 2011, pp. 109-116, doi: 10.1109/CVPRW.2011.5981829. [cited by applicant]
Feldman et al., “Parallel convolutional processing using an integrated photonic tensor core,” https://www.nature.com/articles/s41586-020-03070-1#citeas), Nature, vol. 589, Jan. 7, 2021, 10 pgs. [cited by applicant]
Horst et al., “Cascaded Mach-Zehnder wavelength filters in silicon photonics for low loss and flat pass-band WDM (de-)multiplexing,” https://www.osapublishing.org/oe/abstract.cfm?uri=oe-21-10-11652, 7 pgs., © 2013 Optic… [cited by applicant]
Horst et al., “Compact tunable FIR dispersion compensator in SiON technology,” in IEEE Photonics Technology Letters, vol. 15, No. 11, pp. 1570-1572, Nov. 2003, doi: 10.1109/LPT.2003.818671. [cited by applicant]
Jinguji, K., “Synthesis of Coherent Two-Port Optical Delay-Line Circuit with Ring Waveguides,” Journal of Lightwave Technology, 14(8), pp. 1882-1898, Aug. 1996, 17 pgs. [cited by applicant]
Mehrabian et al., “PCNNA: A Photonic Convolutional Neural Network Accelerator,” https://arxiv.org/abs/1807.08792, © 2018 IEEE. [cited by applicant]
Mehrabian, et al., “A Winograd-Based Integrated Photonics Accelerator for Convolutional Neural Networks,” in IEEE Journal of Selected Topics in Quantum Electronics, vol. 26, No. 1, pp. 1-12, Jan.-Feb. 2020, Art No. 6100… [cited by applicant]
Morcel et al., “FeatherNet: An Accelerated Convolutional Neural Network Design for Resource-constrained FPGAs,” ACM Transactions on Reconfigurable Technology and Systems, vol. 12, Issue 2, Jun. 2019 Article No. 6 pp. 1-… [cited by applicant]
Reuther, et al., “Survey and Benchmarking of Machine Learning Accelerators,” 2019 IEEE High Performance Extreme Computing Conference (HPEC), 2019, pp. 1-9, doi: 10.1109/HPEC.2019.8916327. [cited by applicant]
Unknown, “Blurring an Image,” https://developer.apple.com/documentation/accelerate/blurring_an_image, printed Oct. 15, 2021, 10 pgs. [cited by applicant]
Cited By (1)
US 12,737,648