Product Yield, Test and Diagnostics - Optimizing Diagnosis Workload with Dynamic Partitioning-Based Scan Diagnosis
Product Yield, Test and Diagnostics - Optimizing Diagnosis Workload with Dynamic Partitioning-Based Scan Diagnosis
Wednesday, November 19, 2025
Summary:
During early yield and product ramp, quickly identifying chain and logic defects is critical. As such, diagnosis driven yield analysis has been widely adopted in the industry. The result of diagnosis is a set of logical and physical locations with possible defect types that localize the possible defect area on the failing die. Advanced nodes and features like backside power delivery have made fault isolation extremely challenging, and as such, the ability to process through large volumes of failure information for potential defects more quickly is paramount. In this paper we will present a new technology which demonstrates higher throughput of diagnosis with advancements in memory and runtime efficiency. The experimental results show impressive improvement with memory usage reduced by up to 47%, diagnosis runtime reduced by as much as 3.4x, and resource utilization increasing from 50% to 100%. Dynamic Partitioning partitions the design netlist into smaller, failure-specific subnetlists, it tailors the diagnostic workload to focus only on relevant areas. In summary, Dynamic Partitioning significantly improves scan diagnosis throughput and efficiency, especially during the early yield and product ramp phases, and would be a preferred solution for large scale designs.
During early yield and product ramp, quickly identifying chain and logic defects is critical. As such, diagnosis driven yield analysis has been widely adopted in the industry. The result of diagnosis is a set of logical and physical locations with possible defect types that localize the possible defect area on the failing die. Advanced nodes and features like backside power delivery have made fault isolation extremely challenging, and as such, the ability to process through large volumes of failure information for potential defects more quickly is paramount. In this paper we will present a new technology which demonstrates higher throughput of diagnosis with advancements in memory and runtime efficiency. The experimental results show impressive improvement with memory usage reduced by up to 47%, diagnosis runtime reduced by as much as 3.4x, and resource utilization increasing from 50% to 100%. Dynamic Partitioning partitions the design netlist into smaller, failure-specific subnetlists, it tailors the diagnostic workload to focus only on relevant areas. In summary, Dynamic Partitioning significantly improves scan diagnosis throughput and efficiency, especially during the early yield and product ramp phases, and would be a preferred solution for large scale designs.