aiqbench analysis
GRAU introduces a reconfigurable activation unit that drastically cuts LUT consumption by over 90% for edge accelerators using piecewise linear fitting. By relying on basic comparators and shifters, the design supports mixed-precision quantization and nonlinear functions like SiLU with significantly improved hardware efficiency. While optimal performance is found with 6-8 segments, the architecture highlights a critical trade-off where aggressive low-cost configurations may incur accuracy degradation on complex nonlinearities.
GRAU: Generic Reconfigurable Activation Unit Design for Neural Network Hardware Accelerators
arXiv:2602.22352v2 Announce Type: replace-cross Abstract: With the continuous growth of neural network scales, low-precision quantization is widely used in edge accelerators. Classic multi-threshold activation hardware requires 2^n thresholds for $n$-bit outputs, causing a rapid increase in hardware cost as precision increases. We propose a reconfigurable activation hardware, GRAU, based on piecewise linear fitting, where the segment slopes are approximated by powers of two. Our design requires only basic comparators and 1-bit right shifters, supporting mixed-precision quantization and nonlinear functions such as SiLU. Compared with multi-threshold activators, GRAU reduces LUT consumption by over 90%, achieving higher hardware efficiency, flexibility, and scalability. The best trade-off is usually achieved with 6-8 segments, while complex nonlinearities under aggressive low-cost settings may suffer larger accuracy degradation.
Abstract:With the continuous growth of neural network scales, low-precision quantization is widely used in edge accelerators. Classic multi-threshold activation hardware requires 2^n thresholds for $n$-bit outputs, causing a rapid increase in hardware cost as precision increases. We propose a reconfigurable activation hardware, GRAU, based on piecewise linear fitting, where the segment slopes are approximated by powers of two. Our design requires only basic comparators and 1-bit right shifters, supporting mixed-precision quantization and nonlinear functions such as SiLU. Compared with multi-threshold activators, GRAU reduces LUT consumption by over 90%, achieving higher hardware efficiency, flexibility, and scalability. The best trade-off is usually achieved with 6-8 segments, while complex nonlinearities under aggressive low-cost settings may suffer larger accuracy degradation.
Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI)
arXiv:2602.22352 [cs.AR]
(or arXiv:2602.22352v2 [cs.AR] for this version)
https://doi.org/10.48550/arXiv.2602.22352 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 211 words
