Logo Akiprop

Fast and efficient analog processor for optical data links

Application of energy-efficient high-speed electronic analog arithmetic circuits in artificial neural networks in transmitters and receivers of optical data links for the compensation of linear and non-linear distortions.

Artificial neural networks (ANN) can be used in transmitters and receivers of optical data links to compensate for linear and non-linear distortions on the optical fibre. Compared to conventional signal processing methods, this approach promises higher quality, stability and flexibility. However, the computation of ANNs with classical digital arithmetic units requires large chip area and causes high energy consumption. Therefore, it is crucial for the development of efficient optical data transmission networks to increase the energy efficiency of the electronic signal processing in the highly integrated transmitter and receiver ASICs by means of new circuit techniques.

Optical data link with analog ANN in the receiver to compensate for the chromatic dispersion of the optical fibre.
Optical data link with analog ANN in the receiver to compensate for the chromatic dispersion of the optical fibre.

Integrated analog arithmetic units for ANN applications offer the potential of significantly higher energy and area efficiency compared to classical binary arithmetic units. However, in analogue circuits, computational accuracy is degraded by noise, nonlinearities and mismatch. Furthermore, traditional analogue signal processing circuits in class A operation are not very energy efficient. In this research project, new concepts for energy-efficient high-speed analog arithmetic circuits in an advanced FDSOI CMOS technology are explored and adapted for broadband optical data links.

First, the performance, energy efficiency, accuracy, stability and possible degree of integration of different analog arithmetic circuit concepts will be investigated and evaluated. Then, a specific analogue ANN equaliser circuit for a specific partial function of a transmit or receive ASIC in an optical data link is designed, fabricated and characterized. Finally, the developed circuit is evaluated in terms of its applicability, scalability and performance in ANN-based transmitters and receivers for optical data networks and compared with digital approaches. 

This project is funded by the BMBF under the funding code 16KIS1313
This project is funded by the BMBF under the funding code 16KIS1313

Publications

  1. 2024

    1. J. Finkbeiner, R. Nägele, M. Grözing, M. Berroth, and G. Rademacher, “Characterization of a Femtojoule Voltage-to-Time Converter with Rectified Linear Unit Characteristic for Analog Neural Network Inference Accelerators,” in IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS), 2024, p. accepted.
    2. R. Nägele, J. Finkbeiner, M. Grözing, M. Berroth, and G. Rademacher, “Characterization of an Analog MAC Cell with Multi-Bit Resolution for AI Inference Accelerators,” in IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS), 2024, p. accepted.
  2. 2023

    1. R. Nägele, J. Finkbeiner, V. Stadtlander, M. Grözing, and M. Berroth, “Analog Multiply-Accumulate Cell with Multi-Bit Resolution for All-Analog AI Inference Accelerators,” IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 2023, pp. 1--13, 2023.
  3. 2022

    1. J. Finkbeiner, R. Nägele, M. Berroth, and M. Grözing, “Design of an Energy Efficient Voltage-to-Time Converter with Rectified Linear Unit Characteristics for Artificial Neural Networks,” in IEEE International New Circuits and Systems Conference (NEWCAS), 2022, pp. 327--331.
    2. R. Nägele, J. Finkbeiner, M. Berroth, and M. Grözing, “Design of an Energy Efficient Analog Two-Quadrant Multiplier Cell Operating in Weak Inversion,” in IEEE International New Circuits and Systems Conference (NEWCAS), 2022, pp. 5--9.

Contact

Jakob Finkbeiner

M. Sc.

Research staff member

Raphael Nägele

M. Sc.

Research staff member

To the top of the page