Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Self-calibrated Microring Weight Function for Neuromorphic Optical Computing

J. Garcia-Echeverria, D. Musat, A. Mahsafar, K. R. Mojaver, D. Rolston, G. Cowan, O. Liboiron-Ladouceur

Abstract

This paper presents a microring resonator-based weight function for neuromorphic photonic applications achieving a record-high precision of 11.3 bits and accuracy of 9.3 bits for 2 Gbps input optical signals. The system employs an all-analog self-referenced proportional-integral-derivative (PID) controller to perform real-time temperature stabilization within a range of up to 60 degree Celsius. A self-calibrated weight function is demonstrated for a range of 6 degree Celsius with a single initial calibration and minimal accuracy and precision degradation. By monitoring the through and drop ports of the microring with variable gain transimpedance amplifiers, accurate and precise weight adjustment is achieved, ensuring optimal performance and reliability. These findings underscore the system's robustness to dynamic thermal environments, highlighting the potential for high-speed reconfigurable analog photonic networks.

Self-calibrated Microring Weight Function for Neuromorphic Optical Computing

Abstract

This paper presents a microring resonator-based weight function for neuromorphic photonic applications achieving a record-high precision of 11.3 bits and accuracy of 9.3 bits for 2 Gbps input optical signals. The system employs an all-analog self-referenced proportional-integral-derivative (PID) controller to perform real-time temperature stabilization within a range of up to 60 degree Celsius. A self-calibrated weight function is demonstrated for a range of 6 degree Celsius with a single initial calibration and minimal accuracy and precision degradation. By monitoring the through and drop ports of the microring with variable gain transimpedance amplifiers, accurate and precise weight adjustment is achieved, ensuring optimal performance and reliability. These findings underscore the system's robustness to dynamic thermal environments, highlighting the potential for high-speed reconfigurable analog photonic networks.
Paper Structure (12 sections, 7 equations, 15 figures, 1 table)

This paper contains 12 sections, 7 equations, 15 figures, 1 table.

Table of Contents

  1. Introduction
  2. Weight Function and Controller Design
  3. Weight Function
  4. Microring Resonator Design
  5. Thermal Tuning Operation
  6. Weight Function Control
  7. Calibration Method
  8. Experimental Results
  9. Thermal Tuning
  10. Weight Function
  11. Discussion
  12. Conclusion

Figures (15)

  • Figure 1: Weighted addition function with a single microring and one wavelength per waveguide represented as a) a generic building block of a weighted addition function, b) a single-ended microring-based weighted addition function, and c) a differential microring based weighted addition function.
  • Figure 2: a) Schematic, cross-section, and micrograph of the photonic circuit with double bus MRR and through/drop monitor photodetectors (MPD) to implement tuning, stabilization and weight function with thermal feedback circuit, b) measured transmission spectrum of the MRR, and c) measured thermal shift efficiency.
  • Figure 3: Optical transmission spectrum versus detuning wavelength of a double bus MRR with through, drop and the through minus drop (error signal) curves. The wavelength where the error signal equals to zero is represented by $\lambda_{error}$ and determines the zero value for detuning wavelength axis.
  • Figure 4: Block diagram of the thermal feedback stabilization circuit using through and drop monitoring photodetectors, transimpedance amplifiers, subtractor and proportional-integral-derivative controller that drives the integrated heater of the MRR.
  • Figure 5: Redshift of a fabricated MRR optical spectrum with power applied to the integrated heater to achieve laser wavelength tuning at equal through and drop intensities.
  • ...and 10 more figures