Dataset development of GPU block using Scan Dump for Silicon Debug
Karthik S Kulkarni, H. V. Ravish Aradhya · 2021
In VLSI industry, Si debug is a crucial step of design flow before chipsets reach end users. It is a challenging phase, especially for complex systems like Microcontrollers and Processors which have strong area, power, and timing constraints. Determining the root cause of the issue is the primary step in debug which is expected to be accurate and fast. Various methodologies are followed by industries to trace the root of issue which can be static or dynamic. Later, this issue may be reproduced based on the methodology followed. Scandump, Quick Error Detection (QED), Instruction Footprint Recording and Analysis (IFRA), Node Observability and Traceability are some of the prominent debug methodologies currently used in industries. SoC designs can be implemented on emulation platforms such as Veloce Strato. This paper provides an overview of these methodologies, along with their advantages and limitations. It also provides detailed implementation of Scan dump on GPU that resulted in 100% debug observability.