High performance hardware architecture for multi-mode 1-D forward transform of HEVC

This paper suggests an HEVC 1-D forward transform hardware architecture that supports four TU sizes (4×4, 8×8, 16×16, and 32×32) and yields high throughput. In the proposed transform hardware architecture, coefficient multiplication is required for all TU sizes using a common computation unit equipped with a shifter and adder. Based on the synthesis results using the TSMC 180nm CMOS process and on video of QFHD, maximum operation frequency is 400MHz. The proposed transform can achieve a high throughput rate of 10-Gpels/cycle with 159k of gate area. In addition, it treats all the TU sizes of 4×4, 8×8, 16×16, and 32×32 in 38 cycles equally.