3U-EdgeAI: Ultra-Low Memory Training, Ultra-Low Bitwidth Quantization, and Ultra-Low Latency Acceleration
- Award ID(s):
- 1817037
- Publication Date:
- NSF-PAR ID:
- 10310714
- Journal Name:
- Great Lakes Symposium on VLSI
- Sponsoring Org:
- National Science Foundation
More Like this
-
The paper presents a novel closed-circuit ultra-compact wind tunnel with an 8:1 contraction ratio and high flow quality. Its key design features include a 2D-type main diffuser, a minimum length contraction, and a screened expanding vane cascade in the last corner. These features reduce the overall tunnel footprint area to less than half of the area of a conventional tunnel design with the same test section size and same contraction ratio. The test section turbulence level is a very low 0.03% over a 4Hz–8500Hz bandwidth.