Publicado em: 15/07/2024

Best Paper Award do IEEE CS ISVLSI 2024

O artigo dos autores Rafael da Silva, Mateus Grellert e Ricardo Reis, recebeu o Best Paper Award do IEEE Computer Society Annual Symposium on VLSI.

O artigo “Energy-Efficient Design of Approximate VVC Interpolation Filters Units“, tendo como autores Rafael da Silva, Mateus Grellert e Ricardo Reis, recebeu o Amar Mukherjee Best Paper Award do IEEE Computer Society Annual Symposium on VLSI, realizado de 1 a 3 de Julho em Knoxville, USA.

O artigo foi apresentado pelo prof. Ricardo Reis, que também foi Keynote Speaker do evento com a palestra “Some Trends in SoC Power Optimization “.

Resumo do artigo premiado:
The Versatile Video Coding (VVC) standard in- troduces an array of innovative tools designed to significantly enhance coding efficiency compared to its predecessor, High Efficiency Video Coding (HEVC). For example, the current draft of the VVC standard introduces a block-level adaptive motion vector resolution (AMVR) scheme for coding motion vector dif- ferences with different precision. Consequently, there’s a growing need to develop dedicated accelerators that efficiently explore parallelism for computationally intensive encoding tasks. This work presents a low-power hardware accelerator for Versatile Video Coding (VVC) interpolation step. The proposed design includes the alternative Gaussian filter within this standard, and we also propose an approximate version of the alternative filter, which sets our design apart from competitors. A data gating mechanism is also implemented to prevent unnecessary circuit switching with data gating activated, and we obtained 5.96% to 10.05% power savings. Experimental results show that it can achieve savings of up to 33.97% in power and 31.54% in gate count by using approximate computing techniques.