Design of an Energy Efficient Voltage-to-Time Converter with Rectified Linear Unit Characteristics for Artificial Neural Networks
- Resource Type
- Conference
- Authors
- Finkbeiner, Jakob; Nagele, Raphael; Grozing, Markus; Berroth, Manfred
- Source
- 2022 20th IEEE Interregional NEWCAS Conference (NEWCAS) Interregional NEWCAS Conference (NEWCAS), 2022 20th IEEE. :327-331 Jun, 2022
- Subject
- Components, Circuits, Devices and Systems
Signal Processing and Analysis
Energy consumption
Conferences
Energy resolution
Artificial neural networks
Machine learning
Computer architecture
Energy efficiency
AI accelerators
analog integrated circuits
artificial neural networks
edge computing
energy efficiency
- Language
Machine learning at the edge offers fast, secure and intelligent signal processing. However, calculations need to be very energy efficient because of the limited power budget. This paper presents the design of an energy efficient voltage-to-time converter circuit in 22 nm FD-SOI CMOS technology. The circuit has a rectified linear unit transfer characteristic and is therefore well suited for analog mixed signal computing architectures for artificial neural networks at the edge. Depending on whether mismatch is compensated or not, the effective resolution for a maximum pulse length of 430 ps is 3.0 b or 6.4 b. The simulated energy consumption is below 3 fJ for every output pulse.