AMD Says Natural Language Can Tackle HW and SW Complexity

AMD officially launched its agentic AI assistant, 'AMD ROSS (ROSS).' 'AMD ROSS (ROSS)' is aimed at the full embedded development lifecycle, covering hardware planning, board design, verification, and deployment. The entire lifecycle is handled through natural-language-based control.

AMD described ROSS as distinct from a simple chatbot. While a simple chatbot has the characteristic of answering questions in words, developers can issue instructions in natural language from their existing development screens, with an example prompt given as, 'Create a controller suitable for this chip.' As a result, AI has the character of a 'directly working AI assistant' that can operate specialized semiconductor design software such as Vivado directly and carry out circuit design and command execution.

AMD said development complexity has increased as hardware, software, and AI have become more closely integrated. In this situation, it said ROSS brings together field engineers' everyday technical skills and specialized expertise.

AMD said it expects ROSS to dramatically shorten the time it takes to turn an idea into a finished product.

In a media briefing, an AMD official said the key to the project lies not in mastering design tools themselves, but in creating an environment where engineers can focus on creative design. The official also said safeguards reflecting the realities of engineering sites have been put in place.

AMD said that because semiconductor schematics are a company's core trade secret, ROSS was designed to work even in an internal closed network environment where the internet is completely blocked. It said this closed-network operation removes concerns about technology leaks and has the effect of eliminating per-use AI costs. It also said it has applied measures to prevent circuit damage caused by AI hallucinations, and that the basis for responses is limited to verified AMD official technical documents.

AMD said it has also expanded the flexibility of how developers can use AI. Initially, it can be used as a simple question-and-answer assistant; once trust has been built, it can be used by approving each line of AI-written code; and at a later stage, it can even be entrusted with an entire design all at once. It added that developers can adjust AI's level of autonomy to whatever extent they want.

In real-world development site verification results, ROSS demonstrated its convenience metrics. 'Timing errors,' caused by mismatches in the arrival times of signals inside semiconductors, are considered one of the biggest challenges for engineers, and ROSS independently analyzed the causes of signal errors at 735 locations. It then corrected the 735 signal errors to 0.

For a skilled engineer, a task converting an algorithm into a circuit that would require two sleepless nights was completed by ROSS in 2 hours. It also generated a more efficient circuit that used 35% less internal semiconductor resources. Even in complex cases involving many interlinked components, such as a driver cockpit control system with 12 cameras installed, ROSS completed a complex system planning task that normally takes several weeks in just one day.

ROSS has also completed preliminary verification for more than 800 global customer companies. It is now available immediately through the official website.

AMD plans to pursue expansion into power optimization features and hardware precision diagnostic tools in the fourth quarter of this year. It also plans to launch a rapid deployment system that adds new features every month and to quickly innovate the development ecosystem.

Source: IT DAILY · Kwon Young-seok
Original: https://www.itdaily.kr/news/articleView.html?idxno=241996

References

• https://www.itdaily.kr/news/articleView.html?idxno=241996