Nvidia Weighs Glass Substrates to Stack More HBM on AI Chips

Nvidia is weighing the adoption of glass substrates, working with German equipment maker SCHMID and other companies, to take the performance of artificial intelligence chips to the next level. The strategy aims to sharply improve data processing speeds by increasing the amount of high-bandwidth memory (HBM) that can be mounted.

SCHMID disclosed in a recent first-half earnings presentation that it is collaborating on glass core substrate equipment with key suppliers in the respective supply chains of Intel, Nvidia and AMD, Investing.com reported on the 23rd. Roland Rettenmeier, chief strategy officer, said metallization of through-glass vias (TGV) remains a technical hurdle and that final customer qualification has yet to be completed.

Glass substrates are a next-generation semiconductor packaging material intended to replace conventional plastic substrates. As AI chips have grown more powerful, plastic substrates have run into technical limits, including weak heat resistance and warping. Glass substrates, by contrast, offer superior flatness and rigidity, raising chip speeds by about 40% while cutting power consumption roughly in half.

A low coefficient of thermal expansion is another strength, leaving the material almost free of deformation even at high temperatures. That matters especially for high-performance AI chips, which generate intense heat. Glass also has favorable dielectric properties and low signal loss, transmitting ultra-high-speed signals more accurately, and it deforms less as its size increases, allowing more chips and HBM to be integrated into a single package.

HBM is memory that stacks multiple DRAM chips vertically to speed up data processing. Current AI chips place graphics processing units (GPUs) and HBM on an intermediate substrate known as a silicon interposer. But silicon interposers are limited in how large they can be made, and they are expensive. Using glass substrates would allow the interposer to be replaced, or GPUs and HBM to be mounted without one. That means the overall package can be made thinner while shortening data transmission distances and maximizing bandwidth.

Nvidia is also focusing on improving GPU performance as AI chips must process more data faster. Nvidia Chief Executive Jensen Huang earlier met Chey Tae-won, chairman of SK Group, to discuss cooperation in next-generation semiconductors including glass substrates, and has actively encouraged TSMC to develop glass substrate technology. That suggests the company is giving serious consideration to adopting glass substrates to boost the performance of its next-generation GPUs.

SCHMID Group has been active in developing and supplying solutions essential to glass substrate production. With demand for high-performance computing (HPC) infrastructure surging in the AI era, SCHMID is expanding its product portfolio to include AI server boards and next-generation packaging.