Verification and Fault Injection Platform Based on MTB Stimulus Generation Method for L2 Deep Market Quote Decoder

With the rapid advancement of quantitative trading technology, the demand for low-latency in Level 2 Deep Market Quote (L2DMQ) Decoding is ever-increasing. The L2DMQ decoder faces increasingly significant challenges in terms of bandwidth and performance. As a result, L2DMQ decoders based on FPGA and ASIC technologies are becoming mainstream solutions. Functional verification and testing of L2DMQ decoders implemented as digital circuits became a challenging task. Given the richness of message template types in L2DMQ and the expanded functionalities of decoders potentially leading to decreased stability in the verification platform, efficient and stable decoder functional verification becomes a complex issue. Furthermore, factors like long-term operation or environments with increased levels of electrostatic electricity or charged particle occurrences can lead to functional faults in the decoding circuit. Thus, error injection has become critically important for testing the reliability of the L2DMQ decoder. This paper investigates the encoding method of L2DMQ, providing a comprehensive complexity analysis. Based on the analysis, a message-template-based (MTB) random stimulus generation method is proposed. Compared to the traditional Message-Length-based (MLB) random stimulus generation method, our proposed MTB method simplifies the reference model of the verification platform and enhances its stability. The experimental results show that the verification platform reduces runtime by 26.69% to 69.29%. Building on this, we target a Universal Verification Methodology (UVM)-based fault-injection and detection platform with MTB random stimulus. Experimental results reveal that the obtained functional coverage reaches up to more than 99% when the number of fault injections ranges from 3750 to 22500. This demonstrates the efficiency of the proposed techniques when adopted using UVM and determines error severity levels through the functional coverage while varying different MTB random stimuli.