Samsung Electronics has deployed artificial intelligence directly into its semiconductor design operations, achieving results such as completing verification work that previously took over a month in just two days. Generative AI is moving beyond simple Q&A to dramatically boost productivity in actual engineering tasks, emerging as a key tool to address the company’s shortage of design personnel.
According to reporting by Chosun Biz on the 12th, Samsung’s System LSI Business first opened Anthropic’s “Claude Code” — an AI coding tool powered by the Claude large language model (LLM) — to software developers in May of this year, before expanding its application to specialized semiconductor development work. Within roughly three months of adoption, concrete efficiency improvements are being reported in areas such as custom system-on-chip (SoC) verification and advanced software development.
The most notable achievement came from a data interconnect structure verification project. Under challenging conditions where a client demanded a new chip architecture and used third-party design IP, the verification environment setup and testing — expected to take over a month — was completed in just two days. The work speed was internally assessed to have improved by approximately 15 times. Although some standardized design materials were missing and the RTL (Register Transfer Level) design data for the DRAM controller was not delivered on schedule, Samsung filled this gap using Claude-based AI. When the available SoC design information, on-chip communication specifications, and verification IP data from an electronic design automation (EDA) vendor were fed into the AI, it automatically located, placed, and connected the necessary verification IP, and generated a virtual verification environment and test scenarios. Notably, for the section lacking the DRAM controller design, the AI connected a virtual block to first inspect critical data paths, even catching initial errors before the actual RTL arrived. Internal assessments indicate that no manual errors occurred despite the complex structure involving 64 intertwined data paths.
Cases have also emerged where AI dramatically narrowed the experience gap between seasoned engineers and new hires. To develop software before a chip is physically produced, the behavior of USB devices such as keyboards and mice must be virtually implemented within an emulator. This task typically takes a month because the basic reference code for data transfer provided by EDA vendors is insufficient; engineers must independently study USB communication standards and build device-specific models from scratch. The System LSI Business assigned this task to a second-year engineer with no prior experience in vibe coding or Claude Code. When the employee input the desired functionality and USB reference code into Claude Code, the AI refined the requirements, proposed an implementation method and code, and even supported revisions. As a result, the keyboard and mouse model creation and functional verification were completed in a single day, followed by the development of an Android OS USB device driver using those models.
These achievements are viewed as a breakthrough for addressing the structural workforce gap facing Samsung’s System LSI Business. The division’s total headcount is estimated by the industry at around 6,000, while Qualcomm — a competitor in the mobile AP and SoC market — had approximately 52,000 employees as of September last year. The organizational scale differs by nearly ninefold. Park Yong-in, President and head of the System LSI Business, stated at a management briefing in June that despite record revenue in the first quarter, an annual loss was unavoidable due to sluggish SoC business performance. Industry observers have long pointed to this workforce disparity as a root cause of the underperformance. The System LSI Business is pursuing a strategy of using AI to drastically reduce time spent on repetitive connection and verification tasks and learning new standards, allowing experienced engineers to focus on high-complexity work while accelerating the growth trajectory of junior staff.
AI adoption is also expanding rapidly across Samsung’s entire DS (Device Solutions) semiconductor division. In March, the company announced it had applied AI to analog and logic chip design, reducing some design times by roughly 50%. Last month, the Memory Business disclosed that it had cut the time required to recalibrate process design kits (PDKs) in response to process changes by over 95%, with the system now in active use.
However, unexpected risks have also surfaced in the process of applying generative AI to actual chip design. In one verification task, when instructed to fix an error, the AI — instead of resolving the root cause — simply changed the error message to a general information message. Other reported incidents include the AI reverting previously completed work alongside a requested rollback of a specific feature, and attempting to modify the actual circuit design code (RTL) when asked only to analyze verification results. These issues are attributed to the LLM’s incomplete grasp of the complex dependencies inherent in hardware design languages.
While software can be patched with updates after release, a semiconductor design flaw is difficult to reverse once mass production begins. Accordingly, Samsung is maintaining control by having humans clearly define the AI’s scope of work and re-verify its output, while expanding AI usage in phases. A semiconductor industry insider commented, “LLM-based agents are fast, but without proper control, they can lead to major accidents.” The insider added, “Ultimately, the direction will be to shorten overall development time by reducing multiple manual steps previously performed by humans, allowing engineers to focus on goal-setting and final verification.”