Task 1.1. Design of flexible and scalable SoCs for accelerating Gillespie’s First Reaction Method
In this task we designed a fully parametric IP core (in VHDL) that can be used flexibly to synthesize on demand multiprocessor SoCs for FPGAs implementing efficiently Gillespie's First Reaction Method Stochastic Simulation Algorithm (FRM-SSA).
In particular our flexible FRM-SSA IP core:
- Can simulate biochemical reaction networks with up to at least 4K reactions, each reaction can be up to the 3rd order (considering all possible stoichiometries). This required the careful design of a fully pipelined, low latency unit for computing efficiently reaction propensities.
- Can support two modes of parallel processing: In the Single Simulation in Parallel (SSIP) mode the reactions of a simulation run are partitioned among the available PEs of the SoC evaluating them in parallel. In the Multiple Simulations in Parallel (MSIP) mode, the available PEs of the core run different simulations (SBML files) in parallel.
- Is built using blocks whose functionality is captured in synthesizable HDL descriptions. These descriptions are fully parametric in terms of the number of Reactions (m) of the biomodel(s), number of Processing Elements in the SoC (N) etc. These HDL descriptions, in the form of soft IP cores, can be supplied with specific desired parameters at synthesis time and used to produce SoC implementation instances of any desirable characteristics for a targeted FPGA.
- Can handle the stochastic simulation of large biomolecular networks with at least m=4K reactions in SoCs with at least N=8 PEs connected to a fast switch using a medium size single FPGA. The SoC architecture can scale up both in terms of number of PEs in the FRM core and cores in the FRM SoC.
- The PEs are fully pipelined and support floating point operations for propensity calculations.
Deliverables (technical report)
D1.1 Parameterized HDL SoC descriptions for the FRM-SSA