AI4Design: A Human-Centred, Cross-Domain Framework for Generative AI in Digital System Design Education

Generative artificial intelligence can produce synthesizable register-transfer-level code, testbenches, and design critiques, but it can also displace the cognitive work through which students learn abstraction, timing reasoning, and verification judgement. This paper asks how such tools can be integrated into undergraduate digital system design without exchanging unassisted competence for assisted artifact fluency. We present AI4Design, a design-science framework comprising an executable tool substrate, a pedagogic orchestration layer, and an assurance and assessment layer. Six principles govern its use, including independently derived executable oracles, progressive student autonomy, visible reasoning, and scheduled unassisted assessment. A fourteen-week course illustrates the framework: AI construction support peaks mid-term and is withdrawn for the capstone, while assessment combines executable checks, process evidence, oral defence, and transfer tasks. A cross-layer analysis identifies where software, circuits and boards, chips, and mechanisms can be coupled through executable interface checks. A descriptive tentask benchmark with two open-weight local models provides a limited technical pilot. The 2B model passed 2/10 tasks at first attempt and 4/10 after one repair; the 0.8B model passed 0/10 and 1/10, respectively. Initial successes occurred only on textual tasks. These results concern artifact production, not learning or crossdomain integration. A prospective multi-cohort, mixed-methods protocol specifies how learning, trust calibration, equity, and staff cost should be evaluated. The contribution is a testable course design and evaluation infrastructure, not evidence of educational efficacy.