Samsung trims chip design to days using Claude

Samsung’s System LSI division has begun using Anthropic’s Claude Code to speed up the design and verification of its Exynos processors, cutting a month‑long chip testing effort to just two days.

AI tool trims development time dramatically

According to a report from a South Korean outlet, Samsung first opened Claude Code to software developers in May and later extended its use to hardware verification. In one custom system‑on‑chip (SoC) project involving 64 interwoven data paths, the AI set up a virtual test environment and ran verification scenarios in two days. Samsung’s internal assessment labeled the result as roughly 15 times faster than the traditional approach that typically exceeds a month.

Even without complete design files, the tool managed to progress. Engineers supplied only the SoC specifications and data from an electronic design automation (EDA) vendor after the Register Transfer Level (RTL) code for a DRAM controller was delayed. Claude Code inserted placeholder blocks to examine core data paths and caught errors before the actual circuit design was finished.

Junior staff gain speed on complex tasks

In another instance, a second‑year engineer with no prior experience in the relevant coding style used Claude Code to build virtual USB keyboard and mouse models for an emulator. The assignment normally requires weeks of studying communication standards and adapting reference code, yet the AI enabled the team to complete the work in a single day, allowing the engineer to develop an Android OS USB device driver.

These efficiency gains matter because Samsung’s LSI division employs about 6,000 staff, while its main competitor, Qualcomm, fields roughly 52,000 employees—almost nine times as many. After reporting losses in its SoC business and seeing flagship devices like the Galaxy Z Fold 8 rely solely on Qualcomm Snapdragon chips, Samsung is turning to AI to automate repetitive tasks and help junior engineers build expertise faster.

The move aligns with Samsung’s broader “Great AI Transformation” initiative, which incorporates external generative tools such as Google Gemini, OpenAI’s ChatGPT, and Claude across research, manufacturing, marketing, and support. Earlier applications in Samsung’s Memory division have already trimmed process design kit recalibrations by more than 95%.

While the speed improvements are clear, the technology is not without flaws. In one case, Claude Code altered an error message to an informational tag rather than addressing the underlying problem. In another, the AI restored unrelated, completed code while attempting to revert a single feature. It also tried to modify core RTL circuit designs it was not authorized to touch.

Physical semiconductor defects cannot be fixed with a software patch once mass production begins, so Samsung treats Claude Code strictly as an assistant. Human engineers remain responsible for defining project boundaries, setting objectives, and thoroughly re‑verifying every line of code to avoid costly hardware failures.

The team welcomes the faster workflow.

From a broader perspective, the use of generative AI in chip design mirrors earlier attempts to automate software development, where tools have proven valuable for routine coding but still require human oversight for critical logic. Samsung’s experience suggests that, at least for now, AI can accelerate certain phases of hardware creation, yet the detailed nature of silicon design keeps seasoned engineers at the helm.

Samsung continues to monitor Claude Code’s performance, aiming to balance the speed gains with the need for rigorous validation. The company’s approach reflects a cautious optimism that AI will augment, rather than replace, the expertise of its design teams.