Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE credits
Modern system-on-chip (SoC) verification increasingly relies on virtual platforms tobegin software development early and to reduce dependence on slow register-transferlevel (RTL) simulation. This thesis investigates whether existing SystemC transaction-levelmodeling (TLM) reference models and subsystem-generated data can be reused directlywithin a System Virtualization Platform (SVP) to verify downlink (DL/Tx) and uplink(UL/Rx) processing chains. The intended benefit is straightforward: known-good inputvectors, configuration data, and expected output data should be consumed at the platformlevel without repeatedly executing the complete subsystem reference simulations.
The work followed an engineering case-study method. The SVP environment wasconfigured, Arm intellectual-property (IP) integration was enabled, and the platform wasinitialized. Reference simulations for both DL andULwereexecuted. FortheDLchain, inputdata, configuration-register data, and expected outputs were exported to files accessiblefrom SVP. For the UL chain, input data and configuration-register data were extracted,whereas intermediate and final expected-output extraction was not performed. A platformlevel DL script was developed, register-write commands targeting common and tile-specificregisters were issued, in-phase and quadrature (IQ) traffic was initiated, and internal datadumping was enabled across several processing stages. The DL platform simulation wascompleted and produced output files. UL SVP execution and numerical comparison werenot performed.
The principal result is not full functional equivalence. Instead, the study identifiesa material gap between the behavior available in hardware-oriented or subsystem-levelsimulation and the functionality represented in the current SVP software model. Althoughthe PCIe/AXI configuration writes were issued without explicit error messages, theirexpected effect on the modeled processing chain was notdemonstrated. Unmappedmemoryregions and unavailable testbench features also limited end-to-end validation, while theconfiguration-path behavior requires further investigation. These findings suggest thatdirect TLM reference-model reuse is technically promising. However, its verification valuedepends on platform completeness, address-map consistency, and explicit modeling of therequired test infrastructure. The thesis contributes a reproducible integration method, adata-reuse workflow, a limitation taxonomy, and a validation plan for completing DL andUL equivalence testing. Performance benefits are discussed as expected outcomes ratherthan claimed measurements because comparative timing data have not yet been collected.Keywords: SystemC, TLM-2.0, virtual platform, ASIC verification, SoC verification, downlink, uplink, reference model, simulation reuse, AXI, IQ data.
2026.