Intel’s ZAM Memory Threatens HBM’s AI Throne With 2x The Bandwidth of HBM4, More Capacity & Low Thermal Constraints

Intel's Z-Angle Memory (ZAM) is approaching completion as it races towards taking a bite at the AI boom while challenging HBM as a viable alternative.

Intel's ZAM Challenges HBM As A Big Memory Innovation In the High-Bandwidth, High-Capacity Segment Offering 2x The Speed of HBM4

Z-Angle Memory or ZAM has been stirring up a lot of talk in the memory segment. The upcoming memory standard is being developed by Intel and SoftBank & aims to offer a low-power, high-density replacement to HBM.

Now, new details have been shared that provide more insight into ZAM memory. For starters, the new memory is set to offer 2x the bandwidth as HBM4, even rivaling the next-generation HBM4E standard, which isn't expected till next year. ZAM itself is aiming at a 2028-2030 timeframe, so it will be some time before the project hits production level.

At VLSI Symposium 2026, Intel and SoftBank subsidiary, SAIMEMORY, will unveil new details and have already previewed some features of the upcoming standard. The new details have highlighted various aspects that we will discuss in detail.

Starting with the design itself, ZAM or Z-Angle memory is being demoed with a 9-layer stacked design. A single stack includes eight DRAM stacks, each with a 3 micron silicon substrate between each stack. The main substrate features a single Logic controller that supplements all nine DRAM stacks.

There are three main TSV layers, with each layer packing 13.7k Through-Silicon-Vias inter-connect paths that utilize hybrid bonding. Each layer offers 1.125 GB per layer, so we are looking at a 10 GB capacity per stack or 30 GB in the entire package. The ZAM stack measures 171mm2 (15.4 x 11.1mm), and the memory offers 0.25 Tb/s bandwidth per mm2 equaling 5.3 TB/s of bandwidth per stack.

Now let's talk about the advantages of HBM memory. Currently, HBM is the preferred choice for high-performance AI accelerators and GPUs. But as HBM scales up, it also leads to structural problems, including more heat and higher power. ZAM addresses three core areas: high-density, wide-bandwidth, and lower power consumption. ZAM's structural characteristics allow for a vertical build that is great for heat dissipation without the need to pass through the wiring layer.

Key Advantages of ZAM:

• Higher Bandwidth Density: ~0.25 Tb/s/mm² (vs. lower in HBM)
• Lower Power Consumption: Optimized for low data transfer power
• Superior Heat Dissipation: Vertical architecture allows better thermal management (HBM suffers from heat buildup due to wiring layers)
• Ultra-High Stacking: Supports 9+ layers with extremely thin 3μm Si per stack and via-in-one TSV
• Innovative Tech: Magnetic field coupled wireless I/O + advanced bonding for scalability
• AI-Optimized: Addresses HBM's structural limits for generative AI workloads

The final goal with ZAM is to achieve a dense 3D memory design through 3.5D packaging technology, which houses vertical and horizontal layers, including the high-bandwidth, large memory stack, Power/Ground rails, Silicon Photonics, and Legacy IO, all on a single substrate. ZAM sure does sound very promising, and we look forward to the first real-time demo, but for it to be recognized as a truly competitive solution against HBM, it needs to see real application, and what better way to do so speedily when the AI markets are going in full force.

News Sources: OGAWA Tadashi , Alex

Follow Wccftech on Google to get more of our news coverage in your feeds.