MemMap-pd: Performance Driven Technology Mapping Algorithm for FPGAs with Embedded Memory Blocks

Abstract

Modern day Field Programmable Gate Arrays (FPGA) include in addition to Look-up Tables, reasonably big configurable Embedded Memory Blocks (EMB) to cater to the on-chip memory requirements of systems/applications mapped on them. While mapping applications on to such FPGAs, some of the EMBs may be left unused. This paper presents a methodology to utilize such unused EMBs as large look-up tables to map multi-output combinational sub-circuits of the application, with depth minimization as the main objective along with area minimization in terms of the number of LUTs used. This paper presents a new algorithm for technology mapping onto heterogeneous architectures containing LUTs and embedded memory blocks. For the first time, the concept of reconvergence is used in the field of FPGA mapping and is shown to be effective. The algorithm consists of four main stages, namely, Pre-Processing, Reconvergence Analysis, Memory Mapping and LUT Mapping. Experimental results show that our proposed methodology, when employed on popular benchmark circuits, leads to upto 14% reduction in depth compared with the DAGMap, along with comparable reduction in area.

Pre-Processing: In the first stage of the algorithm, the given circuit is converted into an equivalent two-input network. It has been shown that this conversion leads to better mapping of the circuit into LUTs by minimizing the overall depth of the decomposed circuit.

Reconvergence Analysis: In this stage, the circuit obtained from the pre-processing stage is analyzed for reconvergence and overlapping reconvergent regions are identified for mapping into embedded memories.

Memory Mapping: We use a 2-phase heuristic for selecting appropriate regions for memory mapping. In the first phase, the overlapping reconvergent regions that can be mapped to the memory blocks are expanded till they just satisfy the pin constraint imposed by the memory arrays. In the next phase, the best among the expanded regions are selected based on the potential depth reduction obtained by mapping the region onto embedded memory blocks.

LUT Mapping: This is the final phase of the algorithm in which the residual circuit left after mapping onto memory blocks is mapped into LUTs. The DAG-Map algorithm is used to implement this mapping.