One paper accepted to IEEE Access (Journal, SCIE)

One paper was accepted to IEEE Access. Our paper presents a scale-out DNN inference accelerator architecture, called SAVector, which equips vectored systolic arrays and two-level on-chip buffer architecture. SAVector can reduce energy-delay-product (EDP) by 52% while exhibiting similar performance compared to a scale-up accelerator architecture.

Gunjae Koo
Gunjae Koo
Associate Professor