Evaluating AI technologies for automated code generation in interconnect development
Sampo Suokuisma · Tampere University Institutional Repository (Tampere University) · 2025
In recent years, artificial intelligence has received significant attention due to its potential to improve efficiency. In engineering, AI is considered particularly promising, as the ability to generate executable code from high-level abstractions can substantially increase the productivity of individual engineers. When such gains are scaled across large engineering organizations, it becomes clear why companies are eager to explore AI adoption. Against this backdrop, this thesis examines and evaluates AI approaches for generating functional code for interconnect components within the context of System-on-Chip development. The thesis aims to examine the capabilities, strengths, and limitations of relevant AI technologies, assessing each option in terms of its maturity and its ability to integrate into existing engineering toolchains. In parallel, the specific SoC development context of the case company is incorporated to ensure that the analysis reflects the organization’s needs and supports its exploration of AI opportunities. Finally, the thesis develops an evaluation framework that the company can use to assess and select AI tools and technologies for future research and development activities. These three objectives aim to answer the main research question: How can AI technologies be evaluated and selected to support the generation of functional code for interconnect components in a strategic engineering context? Literature review at the beginning of the thesis was conducted to establish a comprehensive understanding of both the technological landscape and the organizational factors influencing AI-driven code generation. To understand these dynamics, the literature review examines the foundations of automated code generation, current AI approaches for synthesizing code, and approaches used to assess emerging technologies in structured development environments. In addition, a strategic evaluation framework was developed and later refined based on the insights gathered from the expert interviews. Following the literature review, semi-structured expert interviews were conducted, with the questions aligned to the structure of the previously developed framework. The qualitative data gathered from these interviews was analyzed and organized into four thematic clusters: the current state of AI within the company, technical maturity of AI, trust and governance considerations, and strategic alignment. Lastly, the understanding gained from the literature review was integrated with the empirical findings to produce a comprehensive assessment of AI technologies and their suitability for the organization. The combined results indicate that the suitability of AI technologies for automated interconnect code generation cannot be determined solely by technical performance. Instead, effective adoption requires simultaneous consideration of accuracy, capability and governance requirements. Building on these insights, the developed framework integrates technical, organizational, and strategic perspectives, enabling the case company to systematically compare AI approaches and support evidence-based selection during early development planning. The framework not only guides initial evaluation but also provides a basis for continuous monitoring as AI capabilities and engineering practices evolve. The thesis also highlights several avenues for future research, particularly the need to validate the proposed evaluation framework and integration architecture in real engineering tool-chains. By testing the model in operational environments and extending it across different engineering domains, future studies can build the empirical foundation needed to strengthen its generalizability and practical relevance.