SALSA: systematic logic synthesis of approximate circuits

Approximate computing has emerged as a new design paradigm that exploits the inherent error resilience of a wide range of application domains by allowing hardware implementations to forsake exact Boolean equivalence with algorithmic specifications. A slew of manual design techniques for approximate computing have been proposed in recent years, but very little effort has been devoted to design automation.

We propose SALSA, a Systematic methodology for Automatic Logic Synthesis of Approximate circuits. Given a golden RTL specification of a circuit and a quality constraint that defines the amount of error that may be introduced in the implementation, SALSA synthesizes an approximate version of the circuit that adheres to the pre-specified quality bounds. We make two key contributions: (i) the rigorous formulation of the problem of approximate logic synthesis, enabling the generation of circuits that are correct by construction, and (ii) mapping the problem of approximate synthesis into an equivalent traditional logic synthesis problem, thereby allowing the capabilities of existing synthesis tools to be fully utilized for approximate logic synthesis. In order to achieve these benefits, SALSA encodes the quality constraints using logic functions called Q-functions, and captures the flexibility that they engender as Approximation Don't Cares (ADCs), which are used for circuit simplification using traditional don't care based optimization techniques. We have implemented SALSA using two off-the-shelf logic synthesis tools - SIS and Synopsys Design Compiler. We automatically synthesize approximate circuits ranging from arithmetic building blocks (adders, multipliers, MAC) to entire datapaths (DCT, FIR, IIR, SAD, FFT Butterfly, Euclidean distance), demonstrating scalability and significant improvements in area (1.1X to 1.85X for tight error constraints, and 1.2X to 4.75X for relaxed error constraints) and power (1.15X to 1.75X for tight error constraints, and 1.3X to 5.25X for relaxed error constraints).