Uploaded April 2026 | Updated November 2026, 5 days ago
Speaker:
Max Doblas, High-Performance Algorithms and Hardware Accelerators for Bioinformatics
Abstract:
Sequence alignment remains the primary computational bottleneck in modern genomics, exacerbated by the explosive growth of sequencing data and the increasing complexity of long-read analysis. Current solutions force a fundamental trade-off: software on general-purpose CPUs offers flexibility but lacks throughput, while specialized hardware delivers speed but often suffers from algorithmic rigidity and memory constraints.
In this talk, I present a full-stack hardware-software co-design strategy to bridge this gap. I will introduce QuickEd, a novel bound-and-align algorithm that drastically reduces the computational search space for exact alignment without sacrificing optimality. Moving to architecture, I will discuss GMX, a set of ISA extensions designed to accelerate dynamic programming within the CPU pipeline. Finally, I will present SMX, a heterogeneous architecture that couples a flexible lightweight core with a high-throughput co-processor. This design overcomes the area and power limitations of standard CPUs while maintaining the universality required to support diverse alignment models at scale.
Speaker:
Max Doblas, High-Performance Algorithms and Hardware Accelerators for Bioinformatics
Abstract:
Sequence alignment remains the primary computational bottleneck in modern genomics, exacerbated by the explosive growth of sequencing data and the increasing complexity of long-read analysis. Current solutions force a fundamental trade-off: software on general-purpose CPUs offers flexibility but lacks throughput, while specialized hardware delivers speed but often suffers from algorithmic rigidity and memory constraints.
In this talk, I present a full-stack hardware-software co-design strategy to bridge this gap. I will introduce QuickEd, a novel bound-and-align algorithm that drastically reduces the computational search space for exact alignment without sacrificing optimality. Moving to architecture, I will discuss GMX, a set of ISA extensions designed to accelerate dynamic programming within the CPU pipeline. Finally, I will present SMX, a heterogeneous architecture that couples a flexible lightweight core with a high-throughput co-processor. This design overcomes the area and power limitations of standard CPUs while maintaining the universality required to support diverse alignment models at scale.










