Improved Bioinformatics Processing Unit for Multiple Applications

This paper presents a coarse-grain reconfigurable unit for accelerating multiple widely used bioinformatics algorithms. Our design is a highly efficient, programmable bioinformatics processing unit, called BiCell v2. Based on a specialized multimode multiplier, this unit provides three different working modes, in order to accelerate four popular bioinformatics algorithms: Maximum Likelihood based phylogenetic inference, Needleman-Wunsch, Smith-Waterman, and HMMER in sequence alignment. BiCell v2 supports both single and double precision floating-point computation, which significantly increases the accuracy for bioinformatics algorithm acceleration but retains silicon area efficiency. Making use of this improved processing unit, our platform gives about 10X speedup compared to our previous design in single-precision, and 23.2X speedup comparing with GPGPU in the same precision.