Samsung Electronics has put hard numbers behind a packaging bet it is making for next-generation memory, publishing what it calls the first quantitative evidence that hybrid copper bonding (HCB) manages heat better than the conventional thermo-compression bonding (TCB) used in today's high-bandwidth memory.

The reason this matters comes down to a wall the whole industry is approaching: as memory stacks get taller, heat — not electrical performance — becomes the limiting factor. Samsung's claim is that the way the chips are bonded together is one of the biggest levers for getting that heat out, and it has now tried to prove it with measurements rather than projections.

What Samsung Measured

According to reports from industry sources, a Samsung research team systematically demonstrated HCB's thermal advantage over TCB using multi-scale modeling and actual test-chip experiments under conditions approximating a real server environment — going beyond chip- or package-level simulation to prove the benefit in a practical high-stacking setup. The work, "System-Level Thermal Characterization of Hybrid Cu Bonding HBM with 2.5D Advanced Packaging," was published by the IEEE this month. The team built a physics-based model spanning chip-level microstructures up to the package and server-system levels, then validated it by mounting HCB- and TCB-based HBM test chips alongside an ASIC on a silicon interposer and measuring under air-cooling.

The results showed lower hotspot junction temperatures than TCB, reduced thermal interference between the memory stack and the compute chip beneath it, and — under the same cooling — a higher power budget, leaving more headroom for performance. HCB also cut stack height by more than 15%, thinning the package and easing heat buildup.

Why Heat Is the Wall, and Bonding Is the Lever

HBM works by stacking DRAM dies vertically — 8, 12, now 16 layers — directly beside the AI processor, connected through thousands of microscopic vertical wires, to deliver enormous bandwidth in a small footprint. The catch is that every added layer traps more heat in the middle of the stack, and much of it has to escape downward through the very chip doing the computing. As stacks climb, the package can effectively cook itself, throttling performance — which is why bonding, the method that joins one die to the next, turns out to govern how well that heat gets out.

The conventional method, TCB, joins dies with tiny solder bumps and fills the gaps between them with a non-conductive underfill. That underfill sits squarely in the heat's path and acts like insulation, interrupting the flow and raising thermal resistance. HCB removes the bumps entirely and fuses the copper pads of adjacent dies directly, copper-to-copper, with no gap and no underfill. That direct metal contact opens many more paths for heat to conduct away — and because there are no bumps, the whole stack is more than 15% shorter, a thinner package that traps less heat to begin with. That is the mechanism Samsung's paper set out to quantify, showing the advantage holds up in a realistic high-stack setup, not just an idealized simulation.

Why It Matters for the Roadmap

The findings matter because Samsung plans to adopt HCB starting with 16-layer HBM4E — running it alongside TCB at first, then fully at HBM5 — to manage the heat that worsens as stacks climb to 16 layers and beyond. That choice differentiates its approach from the TCB and MR-MUF (mass reflow-molded underfill) methods rivals such as SK hynix lean on, and gives Samsung published, system-level data to back a roadmap it had previously argued mostly in principle.

The catch, which Samsung acknowledges, is cost and complexity. Hybrid bonding demands near-perfect surface cleanliness and alignment across every die, more expensive equipment, and more cleanroom space — and yield risk that rises as the stack grows taller. Those hurdles are exactly why the industry has researched HCB for years without putting it into volume HBM production, and why Samsung is phasing it in rather than switching overnight. "The predictive design framework from this research will be used for bonding evaluation and thermal optimization in next-generation HPC packaging," the team said.

Frequently Asked Questions

What is hybrid copper bonding?

Hybrid copper bonding (HCB), also called direct or bumpless bonding, is an advanced packaging method that stacks chips by fusing the copper pads on adjacent dies directly to each other, copper-to-copper, with no solder bumps and no underfill material in between. Compared with the conventional thermo-compression bonding (TCB) used in today's high-bandwidth memory — which connects dies with tiny solder bumps surrounded by underfill — HCB allows finer interconnect spacing, a shorter overall stack, and, as Samsung's research argues, better heat dissipation. It is considered essential for stacking HBM to 16 layers and beyond.

Why does HBM have heat problems?

High-bandwidth memory stacks many DRAM dies vertically and places them right next to the processor to maximize speed in a small space. But the more layers you stack, the more heat accumulates inside the stack, and much of that heat must escape downward through the compute chip below. As stacks grow from 8 to 12 to 16 layers and beyond, this trapped heat can raise temperatures enough to throttle performance or threaten reliability, making thermal management one of the central engineering challenges in next-generation HBM.

What is the difference between hybrid bonding and TCB?

The difference is in how adjacent chips are joined. Thermo-compression bonding (TCB) connects dies using small solder bumps and fills the surrounding gaps with a non-conductive underfill material — which obstructs heat flow and adds height. Hybrid copper bonding (HCB) eliminates the bumps and underfill, bonding the copper surfaces of the dies directly together. This direct contact lowers thermal and electrical resistance, opens more paths for heat to escape, and reduces the total stack height by more than 15%, according to Samsung. The tradeoff is that HCB is more expensive and technically demanding to manufacture.

When will Samsung use hybrid bonding in HBM?

Samsung has said it plans to begin adopting hybrid copper bonding with 16-layer HBM4E, using it alongside conventional TCB at first, and then move to full hybrid bonding with the next generation, HBM5. Exact timing depends on yield, cost, and customer qualification, and some analysts expect Samsung's first hybrid-bonded HBM to arrive around the end of the decade. The approach differs from rivals such as SK hynix, which has relied on mass reflow-molded underfill (MR-MUF) and has treated hybrid bonding as a later-stage option.