notesum.ai

Published at November 25

UVLLM: An Automated Universal RTL Verification Framework using LLMs

cs.AR

Released Date: November 25, 2024

Authors: Yuchen Hu¹, Junhao Ye¹, Ke Xu¹, Jialin Sun¹, Shiyue Zhang¹, Xinyao Jiao¹, Dingrong Pan¹, Jie Zhou¹, Ning Wang², Weiwei Shan¹, Xinwei Fang³, Xi Wang¹, Nan Guan², Zhe Jiang¹

Aff.: ¹National Center of Technology Innovation for EDA, Southeast University, China; ²Department of Computer Science, City University of Hong Kong, Hong Kong; ³Department of Computer Science, University of York, UK

Arxiv: http://arxiv.org/abs/2411.16238v1

Types	Pre-processing¹		Repair in MS Mode		Repair in SL Mode		UVLLM²		MEIC [17]		Speedup
Types	$FR/\%$	$T_{exec}/s$	$FR/\%$	$T_{exec}/s$	$FR/\%$	$T_{exec}/s$	$FR/\%$	$T_{exec}/s$	$FR/\%$	$T_{exec}/s$	Speedup
Arithmetic³ s⁴	69.93	8.30	13.07	5.60	1.31	0.30	84.31	14.20	62.30	197.29	13.89x
Control s	80.91	7.23	8.18	3.38	0.00	0.00	89.09	10.61	63.64	129.02	12.61x
Memory s	60.00	10.29	28.33	5.14	0.00	0.00	88.33	15.43	54.55	147.17	9.53x
Miscellaneous s	79.65	8.31	7.67	4.77	1.18	0.39	88.50	13.47	66.67	62.67	4.65x
Syntax	74.72	8.49	11.29	5.06	0.98	0.27	86.99	13.83	62.99	134.95	9.76x
Arithmetic f	30.26	3.63	33.33	11.27	2.63	0.64	66.23	15.54	40.53	257.28	16.56x
Control f	29.93	3.33	30.61	9.37	5.44	0.84	65.99	13.54	10.91	163.96	12.55x
Memory f	25.00	4.35	58.33	11.22	3.33	0.87	86.67	16.44	22.73	256.49	15.60x
Miscellaneous f	21.25	3.86	49.06	11.54	5.63	0.90	75.94	16.30	40.07	65.37	4.01x
Function	25.96	3.82	41.46	11.19	4.50	0.78	71.92	15.79	34.57	191.76	12.14x
Overall	51.27	6.16	25.80	7.79	2.68	0.49	79.75	14.77	52.14	153.84	10.42x