Aarti Gupta

Yi Li, Aarti Gupta, Sharad Malik

Specialized hardware accelerators have been extensively used for Deep Neural Networks (DNNs) to provide power/performance benefits. These accelerators contain specialized hardware that supports DNN operators, and scratchpad memory for storing the tensor operands. Often, the size of the scratchpad is insufficient to store all the tensors needed for the computation, and additional data accesses are needed to move tensors back and forth from host memory during the computation with significant power/performance overhead. The volume of these additional data accesses depends on the operator schedule, and memory allocation (specific locations selected for the tensors in the scratchpad). We propose an optimization framework, named COSMA, for mapping DNNs to an accelerator that finds the optimal operator schedule, memory allocation and tensor replacement that minimizes the additional data accesses. COSMA provides an Integer Linear Programming (ILP) formulation to generate the optimal solution for mapping a DNN to the accelerator for a given scratchpad size. We demonstrate that, using an off-the-shelf ILP solver, COSMA obtains the optimal solution in seconds for a wide-range of state-of-the-art DNNs for different applications. Further, it out-performs existing methods by reducing on average 84% of the non-compulsory data accesses. We further propose a divide-and-conquer heuristic to scale up to certain complex DNNs generated by Neural Architecture Search, and this heuristic solution reduces on average 85% data accesses compared with other works.

M V Achutha Kiran Kumar, Aarti Gupta, S S Bindumadhava

Increasing design complexity driven by feature and performance requirements and the Time to Market (TTM) constraints force a faster design and validation closure. This in turn enforces novel ways of identifying and debugging behavioral inconsistencies early in the design cycle. Addition of incremental features and timing fixes may alter the legacy design behavior and would inadvertently result in undesirable bugs. The most common method of verifying the correctness of the changed design is to run a dynamic regression test suite before and after the intended changes and compare the results, a method which is not exhaustive. Modern Formal Verification (FV) techniques involving new methods of proving Sequential Hardware Equivalence enabled a new set of solutions for the given problem, with complete coverage guarantee. Formal Equivalence can be applied for proving functional integrity after design changes resulting from a wide variety of reasons, ranging from simple pipeline optimizations to complex logic redistributions. We present here our experience of successfully applying the RTL to RTL (RTL2RTL) Formal Verification across a wide spectrum of problems on a Graphics design. The RTL2RTL FV enabled checking the design sanity in a very short time, thus enabling faster and safer design churn. The techniques presented in this paper are applicable to any complex hardware design.