Neuromorphic Computing Using Analog-KAN in Standard CMOS Technology Conference

Madanayake, Arjuna, Gadea, Jaime L, Mandal, Soumyajit. (2025). Neuromorphic Computing Using Analog-KAN in Standard CMOS Technology . 00 1006-1010. 10.1109/MWSCAS53549.2025.11244312

cited authors

  • Madanayake, Arjuna; Gadea, Jaime L; Mandal, Soumyajit

authors

abstract

  • Kolmogorov-Arnold Networks (KANs) offer very interesting and largely unexplored architectural advantages compared to deep perceptron systems for artificial intelligence (AI) systems. This paper shows that analog KANs can be efficiently implemented in standard CMOS technology by using multiinput multi-transistor amplifiers to compute analog dot products. Simulation results in a 65 nm CMOS process suggest that such KANs can provide comparable accuracy to multi-layer perceptrons (MLPs) while using fewer transistors per computation.

publication date

  • August 13, 2025

keywords

  • 40 Engineering
  • 4018 Nanotechnology
  • 46 Information and Computing Sciences
  • Machine Learning and Artificial Intelligence
  • Networking and Information Technology R&D (NITRD)

Conference

  • 2025 IEEE 68th International Midwest Symposium on Circuits and Systems (MWSCAS)

publisher

  • Institute of Electrical and Electronics Engineers (IEEE)

start page

  • 1006

end page

  • 1010

volume

  • 00