Simulering av miljoner grindar med Count Algoritmen
DatorsystemSimulationCount-algorithmGate-levelInterpretiveEvent-drivenHierarchicalMulti-queueComputer-aided designLogic evaluationZero-delayDatorsystem
A key part in the development and verification of digital systems is simulation. But hardware simulators are expensive, and software simulation is not fast enough for designs with a large number of gates. As today?s digital zesigns constantly grow in size (number of gates), and that trend shows no signs to end, faster simulators handling millions of gates are needed. We investigate how to create a software gate-level simulator able to simulate a high number of gates fast. This involves a trade-off between memory requirement and speed. A compact netlist representation can utilize cache memories more efficient but requires more work to interpret, while high memory requirements can limit the performance to the speed of main memory. We have selected the Counting Algorithm to implement the experimental simulator MICA. The main reasons for this choice is the compact way in which gates can be stored, but still be evaluated in a simple and standard way. The report describes the issues and solutions encountered and evaluate the resulting simulator. MICA simulates a SPARC architecture processor called Leon. Larger netlists are achieved by simulating several instances of this processor. Simulation of 128 instances is done at a speed of 9 million gates per second using only 3.5MB memory. In MICA this design correspond to 2.5 million gates.