IBM's Copper Replacement Research Is a Power 12 Roadmap Signal

The most useful Power 12 clues are not always processor rumors. Sometimes they are the engineering problems IBM is spending years learning how to solve. IBM's copper-replacement research belongs in that category.

IBM has demonstrated cobalt monosilicide nanowires on 200 mm wafers, advanced ruthenium interconnects below 20 nanometer pitch, coauthored a Cornell-led study of niobium arsenide nanowires, and built a co-packaged optics prototype for AI data movement. Those are confirmed research programs. There is no public evidence that any one of them will appear in Power 12.

Why copper becomes a problem inside the chip

Copper is still excellent at ordinary wire dimensions. The problem appears when chip interconnects narrow below roughly 20 to 30 nanometers. Electron scattering at the wire surfaces and grain boundaries pushes resistance up, while the barrier and liner surrounding copper consume a growing share of the available space.

That is why IBM can take seriously a material whose bulk conductivity does not beat bulk copper. The useful comparison is the finished structure at the scale an advanced chip actually needs. At that point, a liner-free metal or a topological material with surface-dominant conduction may deliver a better result.

The conductor that improves as it shrinks

In June 2026, IBM researchers reported cobalt monosilicide nanowires made with CMOS-compatible processes on 200 mm wafers. The smallest wires had cross-sectional areas of 35 square nanometers, or about 6 nanometers in diameter. Near-epitaxial wires below 100 square nanometers showed roughly an 80 percent resistivity reduction compared with much larger CoSi wires.

The Cornell-led NbAs study published the following month showed the same counterintuitive direction. Single-crystal nanowires reached 40 nanometers in diameter, and their resistivity fell as the wires became smaller. IBM Research in Albany and the IBM Thomas J. Watson Research Center were represented in the author affiliations.

Ruthenium may matter sooner

IBM's ruthenium work looks less like a newly discovered material and more like an advanced manufacturing program. IBM-affiliated researchers demonstrated 16 nanometer pitch subtractive ruthenium lines with resistivity below 20 micro-ohm-centimeters in 2025. Earlier IBM and Samsung work also reported reliability advantages for ruthenium top-via structures with embedded air gaps.

That makes ruthenium the clearest reminder that post-copper does not require one material to win everywhere. IBM could use different conductors at different wiring levels, keep copper where it remains effective, and use optics where an electrical link has become the larger system penalty.

Power systems will care about the distance between chips

IBM's confirmed chiplet direction for the generation after Power 11 makes interconnects more important, not less. Breaking a processor into multiple pieces improves manufacturing flexibility, but those pieces still have to exchange data with low latency and controlled energy use. The links between compute chiplets, I/O, memory, accelerators, boards, and racks become part of the processor architecture.

IBM's co-packaged optics research attacks the longer end of that problem. Its prototype places high-density optical connections close to the package and targets far lower power than mid-range electrical links. IBM has estimated more than a fivefold energy reduction in the link and modeled major AI-training gains when accelerators spend less time waiting for data. Those estimates describe the research prototype, not a Power product.

The Power 12 implication is architectural

It would be easy to turn this into a prediction that Power 12 uses cobalt monosilicide or ruthenium. The evidence does not support that. The more defensible conclusion is that IBM is preparing several answers to the same scaling problem. More compute, more AI acceleration, and more chiplets only help when wiring, data movement, heat, and power delivery scale with them.

Power has always competed on system design rather than a processor core in isolation. That makes IBM's work beyond copper unusually relevant here. A future Power system may gain from these programs through a foundry process, package technology, optical attachment, accelerator connection, or thermal design without putting the name of an experimental material in the product announcement.

The strategic layer is bigger than Power 12

The United States now treats copper dependence and critical-mineral supply chains as industrial-resilience and national-security concerns. The Department of Energy has funded advanced conductor manufacturing, while a 2026 Government Accountability Office assessment called out the difficulty of substituting materials in semiconductors and the need for domestic testing and manufacturing capacity.

Alternative conductors do not remove supply risk by magic. Niobium, cobalt, arsenic, ruthenium, and tantalum all create different sourcing questions. The strategic advantage is having several technically credible options and the domestic knowledge to integrate them. IBM's role spans material discovery, wafer-scale experimentation, packaging, optics, and production hardware. That combination is the real asset.

What would count as real Power 12 evidence

A material moves from roadmap signal to Power 12 evidence only when IBM connects it to the product. That could appear in an announcement letter, Redbook, technical manual, foundry disclosure, packaging presentation, or on-record engineering statement. Before that, the honest label is research with architectural relevance.

The milestones worth watching are 300 mm wafer integration, repeatable thin-film deposition, yield, electromigration lifetime, thermal cycling, contamination control, package assembly, and cost. A conductivity record is interesting. A reliable manufacturing process is what changes a server.

The broader technology, server-farm, cooling, battery, and national-security analysis is published on AS400System.com.