Ideas and Innovation Blog

                                        DuPont Interconnect Solutions

Amid the breathless speculation about their mind-boggling possibilities, advanced computing and artificial intelligence applications face consequential but little-publicized performance hurdles.

Designers must constantly grapple with extreme density on printed circuit boards, ensure signal integrity, and remove unprecedented heat that is generated due to soaring data rates and huge power demands. These hindrances could compromise reliability and stymie the performance of systems.

Each issue must be addressed across the AI board – from logic and high-bandwidth memory (HBM) chips to IC substrates to connectors and more. Failure to manage density, heat and signal challenges across the AI system poses threats to manufacturers’ performance goals.

Despite the obstacles, progress does not stop. While large data sets, powerful algorithms, and robust computing hardware offer significant opportunities, AI applications present AI designers and AI system manufacturers with significant questions:

• How can PCB fabricators quickly and efficiently implement new tools and processes to accommodate the complex design needs of AI chips, now and into the future?

 New wave PCB production technologies are imperative. Transitioning manufacturing equipment and adding innovations in processes will help ensure that the industry can produce ever denser, micron-sized fine-line circuitry that will keep pace with advancements in faster and more powerful HBM and AI processor designs. For example, complex IC substrates that utilize laboratory-engineered and sustainable products born of the material sciences exhibit improved performance and reliability and thus speed the drive toward miniaturization, finer lines and spaces, and greater fine line circuit density.

•  Will dense HBM chips with double or triple the capacity of standard memory chips live up to their promise of   resolving memory bandwidth bottlenecks, yet require less power?

 Because HBM’s vertically stacked memory chips are interconnected by through-silicone vias, HBM exhibits a reduced appetite for energy compared to more power-hungry memory devices. Novel engineered materials can be used to boost the reliability of HBM input-output connections.

•  Will the high-performance thermal interface materials (TIMs), designed to remove unwanted heat from AI processors and HBM chips, also live up to their promise of furnishing the low thermal resistance required to efficiently transfer AI’s larger heat loads, and perform reliably over the long term?

High heat flux and chip flexibility due to CTE mismatches can lessen the effectiveness of advanced thermal materials placed between interfaces, cause microscopic gaps, and lead to potential failure. Advancements in TIMs are lessening this threat. 

What is essential? A transition to enhanced equipment and processes

Paramount to overcoming the immense performance hurdles of AI chips is the implementation of advanced fabrication processes and technologies. First, fabricators must transition to tools that increase accuracy and stability to achieve finer lines and spaces on the board. Second, these manufacturers must embrace materials solutions that can withstand high heat and help enable dense circuitry. Additionally, advanced photoresist and exposure tools are necessary to produce the fine line patterns required by complex AI systems.

For example, DuPont™ Riston® LDI (laser direct imaging) dry film photoresist and Riston® SD2000 (Stepper) dry film photoresist are innovative processes that improve imaging process accuracy and help achieve precision fine-line circuit capabilities within IC substrates. By applying its adhesion and resolution capabilities and chemical resistance, Riston® dry film photoresist can contribute to reduced defect rates and improve production yields. The imaging process must be capable of producing exceptionally fine lines and accurate patterns. The answer comes from utilizing advanced photoresist materials with high dimensional and thermal stability and exposure tools offering high resolution.

Fabrication processes and technologies will continue to evolve, driven by high-density interconnect (HDI) and IC substrate-related innovations. Take line and space precision for example; achieving finer lines and spaces requires extremely precise control over manufacturing processes. A lack of precision controls can jeopardize the integrity of those lines and spaces. When manufacturers apply advanced control mechanisms, they improve the accuracy of exposure tools and the stability of the materials used.

The challenges and promises of HBM chips

HBM chips, which are compact, highly dense, and offer immense memory capabilities, are not new. However, they are essential for the advancement of AI. HBM targets bandwidth-hungry applications. While double data rate (DDR) technology has progressed significantly, HBM’s ultra-high-speed capabilities and power savings have helped it attain prominence. HBM chips include stacked dies, connected through silicon vias, whereas DDR is built on a single, flat memory layout. Thus, HBM delivers higher bandwidth using less power.

Data and instructions storage capabilities of HBM chips allow for much quicker access by AI processors. Complex tasks such as AI processing, data analysis, and multitasking are completed faster.

IC substrates play a primary role in boosting HBM functionality and reliability. They serve as a fine line circuitry platform for input-output memory connections. With lines and spaces becoming smaller and denser and with drill hole sizes narrowing significantly, achieving substrate manufacturing quality and reliability grows in significance.

An innovation enhancing IC substrates is DuPont™ Circuposit™ remover. Since its emergence, it has improved the quality manufacturing of IC substrates and helped deliver consistent process reliability. For instance, Circuposit™ 7840 is a sidewall filler remover. The desmear solution exhibits uniform surface morphology on both conventional and advanced dielectrics. This solution addresses the smear residue left because of the drilling of blind via holes by preventing the smear from insulating the inner layer of copper and the holes’ copper. It also improves peel strength. A surface treatment that bolsters designs, Circuposit™ remover 7800 is used in the desmear process.

Finding answers to excessive AI system heat

One of the most complex design challenges to advancing AI’s capabilities is keeping the chip cool.

Chip designers are encountering tradeoffs. Some materials used to transfer excessive heat offer high thermal dissipation at the start but perform poorly over the long term. Other materials offering a lower heat dissipation rating are nevertheless proving more dependable. Ideally, materials should offer high thermal dissipation at the start and perform reliably over the long term.

Power consumption levels are skyrocketing. Short-term, heat remains the major (and continually growing) challenge. AI processors running at 250 watts today will see the next generation at 500 watts and then further soaring to 1,000 watts, all within a few years.

Escalating wattage makes AI processors the hottest component in the system. Thermal management is being prioritized. Further taxing the TIMs responsible, today’s AI systems have incorporated HBM in close proximity to the main AI processor to reduce latency. This proximity allows designers to cool the HBM chips with the same thermal solution and TIM as the main AI processor.

The proximity of HBM chips to the main processor helps in managing heat more effectively and maintaining system performance. TIMs, especially those using highly engineered fillers and matrices, are playing a key role in manufacturing advanced packaging.

This compares to certain older TIMs. Uneven stress points in chip packages developed during assembly can lead to package warpage. When this occurs, certain older TIMs can develop small cracks, exhibit material pump out, and experience delamination.

An example is Tpcm™ 7000 phase change material from Laird Technologies, a DuPont business. Its thermal conductivity of 7.5 W/mK is designed to enhance the cooling of the most rigorous thermal challenges encountered in advanced computing and AI systems. Tpcm™ 7000 phase change material provides industry leading lowest thermal resistance and a specialty polymeric system offering the most dependable pump-out resistance.

Designers should seek to manage heat by adopting a holistic approach across the entire AI system, instead of component by component. Stronger steps taken toward a holistic approach to thermal management are pivotal in ensuring AI component and system reliability.

Manufacturers must balance initial thermal performance with long-term stability to prevent system failures due to overheating.

 Toward a more holistic view

Advancements in fabrication technologies and foundational IC substrate design are producing manufacturing infrastructures for an increasingly AI-powered world. The continuing goal is achieving a more total solutions view of AI systems, their performance issues, and their capabilities.

As chip designers, process engineers, material suppliers, and more work together, they will uncover solutions that optimize AI system design and improve AI application performance. We collaborate closely with our partners to raise performance and explore and seize every cost- and space-saving advantage of deploying multifunctional products which offer a holistic approach to AI system design. Additionally, we continue to innovate and stay at the forefront of technology, bringing new advancements to our customers to enable the next generation of AI applications.

To successfully enable AI applications, design engineers and manufacturers must take a comprehensive, holistic view of board design, implementing innovative materials solutions across substrates, HBMs, main processors, and more, to tackle the immense reliability challenges that accompany the rise of AI.

Copyright @2025, Laird Technologies, Inc. DuPont™, Laird™, the Laird Logo, and all trademarks and service marks are owned by DuPont de Nemours, Inc., Laird Technologies, Inc., or their