In a monumental development that promises to revolutionize microprocessor design, Intel has introduced innovative glass substrates, signaling a significant departure from the traditional use of silicon. These glass substrates are poised to propel the chip-making industry toward its ambitious goal of accommodating 1 trillion transistors on a single chip by the year 2030.
Substrates, the foundational component of a CPU onto which minuscule ‘traces’ are etched through a process known as photolithography, play a pivotal role in the functionality of microprocessors. Silicon has been the go-to choice for decades, but it has encountered mounting challenges as microprocessors have become denser and metal traces narrower, leading to issues such as warping due to intense heat generation.
Narrower traces enhance CPU speed but hinder heat dissipation, impeding electrical flow and potentially causing silicon deformation as cooling becomes increasingly challenging. Consequently, chip manufacturers have been compelled to explore alternative options as Moore’s Law, the long-standing doctrine predicting that microprocessor technology will double transistor counts every two years, grapples with fundamental physical limitations.
Enter glass substrates, an alternative material championed by Intel.
According to Intel, glass substrates boast the ability to withstand higher temperatures, exhibit “ultra-low flatness” for improved lithographic accuracy, reduce pattern distortion by 50 percent, and offer dimensional stability that enables precise stacking of transistor layers.
This groundbreaking technology, hailed as a potential revival of Moore’s Law, has resulted in “industry-leading glass substrates for advanced packaging” and holds the promise of benefiting key industry players and foundry customers for decades to come, as emphasized by Babak Sabi, Intel’s Senior Vice President and General Manager of Assembly and Test Development.
Utilizing these substrates during CPU packaging, the process of assembling and securing various microprocessor elements, will empower future designs to accommodate an ever-increasing number of transistors while maintaining superior structural stability compared to silicon packaging. Intel’s “viable and essential” glass design will facilitate more precise positioning of traces, enhancing internal communication speeds, significantly boosting performance, and enabling chip-to-chip communication via optical connections.
These capabilities, among others, will enable superior power delivery solutions while achieving high-speed signaling at reduced power levels, ultimately bringing the industry closer to the audacious goal of scaling to 1 trillion transistors on a single package by 2030.
This target represents a significant leap beyond current consumer chips like Apple’s M2 Ultra desktop CPU, which houses a mere 134 billion transistors and employs internal wiring as narrow as 5 nanometers. This extraordinary feat of miniaturization underscores the critical role of efficient chip design in rejuvenating Moore’s Law.
Intel is not the sole player in the development of glass substrate solutions. Japanese company Dai Nippon Printing, for instance, has unveiled a glass core substrate (GCS) solution that provides chip designers with the flexibility to balance thickness, warpage, stiffness, and smoothness.
As the microprocessor market eagerly incorporates this technology into tangible chips (Intel has committed to delivering “complete glass substrate solutions” by the second half of this decade), the new architecture will be indispensable in crafting next-generation processors for applications such as artificial intelligence (AI) and ‘hyperscale’ cloud operations.
Systems built upon these chips, including increasingly intricate designs integrating multiple ‘chiplets’ into a unified package, will hold particular value in the expanding realm of data centers, which are rapidly populating with servers to support the burgeoning use of AI and similar applications.
Australian data center giant Macquarie Data Centers recently announced a substantial expansion of its Macquarie Park Data Center Campus in Sydney, increasing capacity from 10MW to 50MW, making it “the most advanced data center in Sydney”. Additionally, they plan to expand the capacity of their IC3 Super West facility by up to 41 percent, reaching 45MW of computing capacity.
These “advanced AI-ready data centers” are vital for supporting corporate, government, and industrial use of AI, necessitating ever more efficient CPU technology to deliver unprecedented computational power. Furthermore, advanced cooling technologies, both air and liquid-based, are essential to prevent operational overheating.
David Hirst, the Group Executive of Macquarie Data Centers, emphasized the crucial role of AI-ready data centers, stating that “unlocking the full potential of AI hinges on the availability of AI-ready data centers” and that they are designing their upcoming data centers to exceed the demanding requirements for higher densities, ensuring optimal performance in AI model training and inference.
