Samsung introduces Claude to the chip room, and the results show exactly what the artificial intelligence revolution looks like
Samsung is starting to show how artificial intelligence can transform from an auxiliary program into a real working tool in chip design and testing, and the initial results are mind-blowing!
Samsung's LSI Systems unit, which deals with mobile processors and image sensors, among other things, implemented Claude Code in its systems.
The results sound like a distant dream,
In a short time, results that are hard to ignore appeared, tasks planned to take over a month were completed within two days, and in another case, an engineer with about two years of experience completed a development task of over a month in a single day.
The important number here is not just '15 times faster'.
The big story is that Samsung is trying to use AI to solve a structural problem in the chip industry: a shortage of engineers and skilled personnel.
In other words, the figure, 15 times faster, has a deeper meaning.
One of the interesting cases occurred in a project for a custom system-on-chip for a client:
The project was particularly complex, the client demanded a new chip architecture, external design components were used, there were no standard design materials, and a DRAM memory controller needed for testing had not yet arrived.
In the normal process, each of these deficiencies could become a bottleneck.
Samsung fed Claude information about the chip design, internal communication standards, and information about external design components.
The system was able to identify where components were needed for testing, to place and connect them, to create a test environment, and to build virtual test scenarios.
Even when the actual design of the DRAM controller was not yet ready, the AI created virtual blocks that allowed Samsung to test the main data paths in advance.
And this is something very critical to understand:
Instead of waiting for all chip parts to be ready and only then starting to test, the company can move some of the tests to an earlier stage.
In the project, 64 data paths were tested, and the early testing allowed problems to be identified before the full design reached the next stage.
In other words, the AI does not just 'write code'.
It is starting to integrate into the engineering process itself!
One of the amazing figures from the report is that Samsung reported it is trying to close a 9-fold gap in manpower.
Here lies perhaps the greatest business significance:
Samsung's LSI Systems unit reportedly employs about 6,000 people, while Qualcomm employs about 52,000 people.
This is a gap of almost 9 times in organization size.
It is clear that the number of employees of the two companies cannot be directly compared, because Qualcomm operates in a wider range of fields, but in such a market, the gap illustrates Samsung's challenge.
And this is where AI comes in:
If one engineer can complete a task in one workday that previously required more than a month, the company does not necessarily need 30 times more employees to increase development output.
The meaning is increasing the productivity of the existing workforce,
And this is one of the most important stories of the AI revolution, not just replacing employees, but transforming a single engineer into a system capable of performing much of the work around them,
These are, in fact, substantial signs of a revolution.
Instead of asking the system a question and getting an answer, the tool is capable of reading an entire code structure, modifying files, and executing commands.
In chip design, this means using AI for creating test code, building test environments, working with circuit data, test plans, and communication standards.
This transforms AI from something that sits next to the engineer to something that enters the workflow, and that is a huge difference.
But there is also a red flag:
One of the challenges is that AI might make code corrections while hiding errors or create a result that appears correct externally but is not necessarily correct from an engineering perspective.
Therefore, the question is no longer just:
'Can AI do the job?'
But rather:
'How do we know it did it correctly?'
In chip design, this is a critical question, an error in the code can become an error in the design, and a design error can lead to expensive production, delays, and even product failure.
Therefore, the potential of AI is enormous, but so is the need for human oversight.
Samsung does not stop at Claude, the company has also expanded the use of other generative AI tools, including Gemini and ChatGPT, and presented a broader 'AI transformation' plan across the entire organization.
If the move succeeds, the impact could be far beyond saving a few weeks on a single project:
Less time for the engineer, less time for testing, earlier identification of faults, more efficient use of personnel, and shorter development cycles.
In a world where each new generation of chips requires more complexity, more testing, and more personnel, such an improvement can become a real competitive advantage.
And the market? It still mainly looks at how many chips Samsung sells.
Perhaps the more important question is how quickly Samsung can design the next chip.