Samsung Electronics has integrated Anthropic's large language model, Claude, into its semiconductor design and verification processes, dramatically reducing development timelines, according to a report by Chosun Biz on August 12.

Just three months after initially rolling out Claude Code, an AI programming tool, to software developers, the company is already seeing significant efficiency gains on the front lines. In a recent project involving custom System-on-Chip (SoC) verification, a task that was originally expected to take over a month was completed in just two days. Internal assessments indicate that productivity has improved by approximately 15 times.

The project was particularly demanding, as the client introduced a new semiconductor architecture and utilized intellectual property (IP) from an external company. The challenge was compounded by the lack of some standardized design documentation and the delayed delivery of the DRAM controller RTL data required for verification. Engineers fed the available SoC design data, chip internal communication protocols, and verification IP information from the EDA vendor into the Claude model. The AI then automatically located and configured the necessary verification IP, established connections, and generated a virtual verification environment and test scenarios.

For the parts of the DRAM controller design that had not yet been delivered, the AI connected to virtual verification modules, enabling it to prioritize checks on the core data paths. This allowed the team to identify and fix early-stage bugs even before the actual RTL code was delivered. Despite the verification object involving a complex structure with 64 interwoven data paths, the process of building the verification environment avoided the human errors typically associated with repetitive manual operations.

In the semiconductor development workflow, Claude Code is primarily used to generate verification code and build test environments based on circuit design data, verification programs, and communication protocols. Samsung's System LSI division, responsible for designing mobile APs and image sensors, is leveraging Claude to reduce repetitive work and shorten the time needed to master new technologies, thereby significantly boosting the productivity of existing personnel.

Samsung Electronics first made Claude Code available to software developers in May and has since expanded its application to specialized semiconductor development. The company has also broadened its use of other external generative AI models, such as Gemini and ChatGPT, and has formally established an "AI Mega Transformation" strategy covering all related business operations, including R&D, production, marketing, and back-office support.

However, the company's internal evaluations also caution that the scope and output of generative AI must be strictly controlled. In one verification task, an engineer instructed the AI to fix an error. Instead of resolving the root cause, the AI simply changed the "Error Log" into a general "Info Log." In another instance, when asked to revert only a specific feature, the AI reset all other completed work back to its initial state. Furthermore, when asked to analyze verification results, the AI attempted to directly modify the underlying circuit design code (RTL). Internal analysis attributes these issues to the large language model not yet fully grasping the extremely complex dependencies inherent in hardware description languages (HDL).