FLAME RETARDANTS became more aligned with the actual needs of high-end PCB materials.
Experimental validation
The decisive step in any AI-driven materials programme is experimental validation. Predictions must be converted into molecules, which must be tested under relevant conditions. In this program, a substantial set of AI-generated candidates was synthesised and evaluated.
The experimental work served two purposes. Firstly, it tested whether the model could identify molecules with the desired low-loss and FR profile. Secondly, the data improved the model. Each molecule produced new information: which descriptors mattered, which structural motifs were beneficial, which candidates failed, and where the model needed correction.
This created an active learning cycle. The AI proposed candidates; ICL scientists selected and refined them; the laboratory generated real data; and the model learned from the results. In this case, the process ultimately enabled ICL to identify several specific molecules that yielded ultra-low-loss performance and could function as FRs for highend PCB applications.
Why this matters
The irony of this development is that AI is both the market driver and the development tool. AI servers, high-speed switches, advanced accelerators, optical-electrical modules and communications infrastructure require increasingly sophisticated PCB materials.
These systems push signal speeds higher, increase thermal density, and tighten loss budgets. As a result, laminate suppliers and PCB fabricators need dielectric materials with lower Df, stable Dk high reliability and compliant flame retardancy.
At the same time, AI modelling can accelerate the discovery of the very materials needed to support AI hardware. This creates a feedback loop between digital infrastructure and material innovation. Better AI systems require better electronic materials; better AI tools can help discover those materials faster.
The commercial implications are significant. Low-loss laminates are increasingly relevant to data centres, AI accelerators, high-speed networking, telecom infrastructure, radar, automotive electronics and semiconductor packaging. FRs that are compatible with this low-loss regime could therefore occupy a high-value position in the electronics materials supply chain.
The broader lesson
The ICL case illustrates a practical model for AI in chemical innovation. The most effective approach is not a fully autonomous black box. It is a human-guided discovery workflow in which computational tools, chemical AI and experienced scientists work together.
The model can explore a larger chemical space than the laboratory can cover. Computational chemistry can help estimate relevant structureproperty relationships. SBAI can identify patterns and generate new candidates. But expert chemists are needed to judge feasibility, mechanism, synthesis, formulation relevance and commercial fit. Experimental validation remains the final authority.
This is especially important in low-loss FRs, where the target is not a single property. The molecule must satisfy a constrained multidimensional window. It must be FR, low loss, thermally stable, resin-compatible, processable, scalable and commercially viable. AI helps navigate the search, but domain expertise defines what‘ good’ really means.
Conclusion
The development of ultra-low-loss FRs for high-end PCBs represents a new frontier in electronics materials. It is no longer enough for FRs to pass fire tests. In advanced PCB laminates for AI, communications, and highspeed computing, FRs must also preserve signal integrity.
ICL’ s collaboration with NobleAI demonstrates how this challenge can be addressed. By combining computational chemistry, SBAI, molecular generation, expert chemical review, synthesis, testing and iterative model improvement, the team moved from a broad generated set of more than 4,000 molecules to a focused group of experimentally validated candidates. The result was the identification of specific molecules capable of delivering ultra-low-loss performance while functioning as FRs for demanding PCB applications.
This approach points toward the future of speciality chemical development. The winning materials will not come from AI or traditional experimentation alone. They will come from integrated workflows that combine scientific understanding, digital modeling, and practical industrial expertise. For high-end electronics, the goal is clear: fire safety without signal interference. ●
J j
Dr Ronny Costi
DIRECTOR OF GROWTH & INNOVATION
ICL GROUP Ronny. costi @ icl-group. com www. icl-group. com
SEP / OCT 2026 SPECCHEMONLINE. COM
71