Samsung Electronics Introduces Claude Code to Improve Development Efficiency, Reducing Chip Design Validation Time from One Month to 2 Days

August 12th. Since opening Anthropic's AI programming tool Claude Code to software developers in May this year, Samsung Electronics' System LSI Division has used it for customer-specific SoC feature validation and early-stage semiconductor development software. In a recent customer-specific SoC project, Samsung shortened the originally expected over one-month verification environment setup and validation work to 2 days, increasing internal evaluation speed by about 15 times.

The project involved 64 data paths, and some standardized design materials and DRAM controller RTL were not provided promptly. Samsung input existing SoC design information, on-chip communication specifications, and EDA vendor verification IP information into Claude. The AI then completed the positioning, connection, virtual verification environment, and test scenario construction of the verification IP, and discovered early-stage errors using virtual modules before the actual RTL completion.

In another case, an engineer with only two years of experience, using Claude Code without prior relevant experience, completed in one day what usually takes about a month for USB keyboard and mouse model development and validation, and further completed Android USB device driver development. Samsung hopes to increase the productivity of existing design personnel through AI to narrow the manpower gap with competitors; its System LSI Division has about 6,000 employees, while Qualcomm has a total of about 52,000 employees.

However, Samsung has also found that the scope and results of generative AI need to be strictly controlled internally. In some verification tasks, the AI has rewritten error messages as general messages to avoid error reporting, mistakenly rolled back other completed work, and even attempted to modify actual RTL design code when only analyzing verification results. Therefore, Samsung still requires engineers to define the scope of AI operations and manually review the results.