A Prototype-Based Gate-Level Cycle-Accurate Methodology for SoC Performance Exploration and Estimation

A prototype-based SoC performance estimation methodology was proposed for consumer electronics design. Traditionally, prototypes are usually used in system verification before SoC tapeout, which is without accurate SoC performance exploration and estimation. This paper attempted to carefully model the SoC prototype as a performance estimator and explore the environment of SoC performance. The prototype met the gate-level cycle-accurate requirement, which covered the effect of embedded processor, on-chip bus structure, IP design, embedded OS, GUI systems, and application programs. The prototype configuration, chip post-layout simulation result, and the measured parameters of SoC prototypes were merged to model a target SoC design. The system performance was examined according to the proposed estimation models, the profiling result of the application programs ported on prototypes, and the timing parameters from the post-layout simulation of the target SoC. The experimental result showed that the proposed method was accompanied with only an average of 2.08% of error for an MPEG-4 decoder SoC at simple profile level 2 specifications. 1. Introduction Due to the advancement of the semiconductor processing technology, the system-on-a-chip (SoC) design is more popular in consumer electronics than it ever will be. SoC design can effectively reduce system price and promote more functions in a compact product. The embedded processor core, coprocessor, memory subsystem, on-chip bus, intellectual properties (IPs), and various I/O peripherals are well organized on an SoC hardware platform to perform complex functions. In modern products, hierarchal software architecture yields boot loader, embedded OS, graphic user interface (GUI), IP device driver, and application software, which all operate on an SoC. However, the system level of hardware/software (HW/SW) cosimulation and coverification tools are limited by the computational limit of the simulation platform and the tradeoff of model accuracy. The adjustment of simulation time versus exactness always harms system designers, especially communication and standard compatible products with huge simulations. Prototype simulation plays the role of final HW/SW function verification. There are several HW/SW codesign EDA tools in the design phase, such as Mentor Graphics Seamless [1], Synopsys Eaglei [2], and CoCentric [3]. Although these EDA tools allow HW/SW cosimulation at bus level, where each bus transaction involves all signals necessary to complete the bus function, the simulation efficiency is restricted to the