A 3D Tiled Low Power Accelerator for Convolutional Neural Network

Huan, YuxiangKTH,Skolan för elektroteknik och datavetenskap (EECS),Fudan Univ, State Key Lab ASIC & Syst, Shanghai, Peoples R China. (författare)

QC 20181210
It remains a challenge to run Deep Learning in devices with stringent power budget in the Internet-of-Things. This paper presents a low-power accelerator for processing Convolutional Neural Networks on the embedded devices. The power reduction is realized by exploring data reuse in three different aspects, with regards to convolution, filter and input features. A systolic-like data flow is proposed and applied to rows of Processing Elements (PEs), which facilitate reusing the data during convolution. Reuse of input features and filters is achieved by arranging the PE array in a 3D tiled architecture, whose dimension is 3 x 14 x 4. Local storage within PEs is therefore reduced and only cost 17.75 kB, which is 20% of the state-of-the-art. With dedicated delay chains in each PE, this accelerator is reconfigurable to suit various parameter settings of convolutional layers. Evaluated in UMC 65 nm low leakage process, the accelerator can reach a peak performance of 84 GOPS and consume only 136 mW at 250 Mhz.

Xu, JiaweiFudan Univ, State Key Lab ASIC & Syst, Shanghai, Peoples R China. (författare)
Zheng, Li-rongKTH,Fudan Univ, State Key Lab ASIC & Syst, Shanghai, Peoples R China.(Swepub:kth)u1q4gmpy (författare)
Tenhunen, HannuKTH,Integrerade komponenter och kretsar(Swepub:kth)u1wjjaxp (författare)
Zou, ZhuoFudan Univ, State Key Lab ASIC & Syst, Shanghai, Peoples R China. (författare)
KTHSkolan för elektroteknik och datavetenskap (EECS) (creator_code:org_t)

Ingår i:2018 IEEE INTERNATIONAL SYMPOSIUM ON CIRCUITS AND SYSTEMS (ISCAS): IEEE9781538648810

Av författaren/redakt...: Huan, Yuxiang; Xu, Jiawei; Zheng, Li-rong; Tenhunen, Hannu; Zou, Zhuo

Om ämnet

Kungliga biblioteket hanterar dina personuppgifter i enlighet med EU:s dataskyddsförordning (2018), GDPR. Läs mer om hur det funkar här.
Så här hanterar KB dina uppgifter vid användning av denna tjänst.