Intel debuts glass core substrate for multi chip accelerators

Intel has introduced a 10-2-10 glass core substrate with EMIB multi chip connectivity, targeting next generation high performance Artificial Intelligence and HPC accelerators with larger silicon area and finer signal routing.

At NEPCON Japan 2026, Intel showcased a new glass substrate technology aimed at powering the next generation of advanced Artificial Intelligence and high performance computing accelerators. According to SemiVision Taiwan, Intel has created a high performance core glass substrate measuring 78×77 mm for the entire package. This substrate can accommodate approximately twice the reticle size with silicon, equating to about 1,716 mm² (2×858 mm²) of silicon area for logic and memory.

Intel describes this as the first 10-2-10-thick glass core substrate that integrates its EMIB multi chip module technology. The 10-2-10 designation refers to 10 redistribution layer build up layers on the top of the substrate, a 2 layer glass core, and 10 build up layers on the bottom side. The 10 redistribution layer (RDL) build up layers on the top of the substrate are used to redistribute signals from the die, and these layers handle fine pitch routing, with Intel dedicating as many as 10 layers for this purpose.

The core of the glass substrate consists of two layers and is made of 800 μm (or 0.8 mm) class material, which may include embedded metal layers for through-glass-vias or power and ground planes. The final ’10’ number denotes the number of build up layers on the bottom side, mirroring the top for symmetric routing to connect to the motherboard or PCB, and this bottom stack helps declutter the dense wiring from the silicon die so it can interface more cleanly with a standard board. Intel’s 10-2-10 glass core substrate is also equipped with some of the finest bump pitches in the market, with 45 μm (only 0.045 mm) bump pitch. Finally, Intel embeds two EMIB bridges for multi-chip-module packaging so you can manufacture several smaller dies and interconnect them via EMIB bridge, present on the glass core.

55

Impact Score

AMD and Rackspace plan dedicated AI compute rollout

AMD and Rackspace have finalized a phased deployment for dedicated AMD-based compute across Rackspace data centers. The capacity is aimed at regulated enterprise workloads, including clinical AI and large-scale inference.

Lexar tests SSD offloading for local AI models

Lexar is developing an AI-focused SSD approach designed to cut DRAM demand when running large language models on consumer PCs. Internal tests show the company’s storage offloading can load models that traditional local frameworks struggle to run with limited memory.

NVIDIA Blackwell leads MLPerf Training 6.0

NVIDIA’s latest MLPerf Training 6.0 results put Blackwell across every benchmark in the suite, including new MoE workloads. Partner systems from Microsoft Azure and CoreWeave highlighted large-cluster runs on Llama 3.1 405B and DeepSeek-V3 671B.

HPE and NVIDIA expand AI Factory for agentic systems

HPE and NVIDIA are adding agent tooling, confidential computing and updated accelerated systems across the HPE AI Factory portfolio. The expansion targets production deployments that need governance, secure data handling and integrated networking.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.