The Wire Congestion Problem
Every transistor in a chip needs two things: signals and power. Traditionally, both are delivered from the same side — the front of the chip, through the metal interconnect stack. As transistor density has increased with each node, the routing congestion for both signals and power has become a serious bottleneck. Power delivery wires take up routing tracks that could otherwise carry signals, and they have to be thick enough to handle the current without excessive voltage drop (IR drop).
At the 3nm node and below, this congestion is acute. The power delivery network (PDN) can consume 20-30% of the metal routing resources on the lower metal layers — the very layers where signal routing is most constrained.
Backside power delivery solves this by moving the entire power network to the back of the wafer, leaving the front side exclusively for signal routing. It's a simple concept with enormously complex execution.
How Backside Power Delivery Works
The Basic Idea
After front-end transistor fabrication and initial metal layers are complete on the front side, the wafer gets bonded face-down to a carrier wafer. The original silicon substrate is then thinned from the back — ground and etched down from 775 µm to just a few micrometers. Through-silicon vias (nano-TSVs) are etched from the back side down to the transistor level. A power delivery metal stack is built on the back side, connecting to the transistors through these nano-TSVs.
The result: signal wires on the front, power wires on the back. Complete separation.
Related reading: RISC-V Custom Extensions: Building Application-Specific Proc.
Intel PowerVia
Intel was first to demonstrate backside power delivery in a product-relevant context. Their PowerVia technology, part of the Intel 20A and 18A process nodes, uses nano-TSVs roughly 30-50nm in diameter to connect the backside power grid to the front-side transistors. Intel showed functional SRAM test chips with PowerVia in 2023 and claimed a 6% frequency improvement and 30% reduction in IR drop compared to front-side power delivery.
The key advantage Intel claims is that PowerVia uses a standard metallization process on the back side — the same deposition and patterning techniques used for regular interconnects, just applied to the wafer's back side after thinning.
TSMC's Approach
TSMC has been more measured in their public statements about backside power delivery. They've presented research results at conferences (IEDM, VLSI Symposium) showing functional test structures with backside PDN, but they haven't committed to a specific production timeline the way Intel has. TSMC's approach reportedly uses a similar nano-TSV concept but may differ in the specific process integration.
Samsung has also demonstrated backside power delivery in their research labs, presenting results at various IEEE conferences. Their timeline appears to be slightly behind Intel's for production deployment.
Related reading: RISC-V vs ARM 2026: Cuộc Chiến Kiến Trúc Chip.
Performance Benefits
Reduced IR Drop
With dedicated backside power routing, the power wires can be made wider and thicker without competing for space with signal wires. This dramatically reduces resistive voltage drop across the chip. Lower IR drop means more consistent supply voltage at the transistors, which translates directly to better timing margins and higher achievable clock frequencies.
Signal Routing Freedom
Freeing up 20-30% of front-side routing resources has cascading benefits. Place-and-route tools can find shorter signal paths, reducing wire delay and power. The reduced congestion also improves routability for complex designs that were previously hitting routing limits. For standard cell libraries, backside power delivery enables simpler cell architectures with fewer internal routing constraints.
Cell Height Reduction
With power rails moved to the back side, standard cell heights can be reduced. TSMC and Intel have both discussed moving from 5-track or 4.3-track cell heights to even more compact configurations with backside PDN. Smaller cells mean higher logic density — effectively getting a density boost without shrinking the transistor itself.
Manufacturing Challenges
Wafer thinning is the scariest part. You're taking a wafer with billions of dollars worth of processing on it and grinding it down to a few micrometers of silicon. Any mishandling, contamination, or stress-induced cracking during this step can destroy the wafer. The carrier wafer bonding must be perfectly uniform — any void or particle at the bonding interface becomes a defect.
Related reading: Semiconductor Supply Chain 2026: Geopolitics, Reshoring, and.
The nano-TSVs must align precisely with the front-side transistor contacts. With overlay budgets in the single-nanometer range, this requires extremely accurate back-side lithography using alignment marks transferred through the thinned silicon.
Thermal management is another concern. The thinned silicon substrate has lower thermal conductivity than the original thick wafer, potentially making heat dissipation harder. However, the backside metal stack can actually help — metal is a good thermal conductor, and the power delivery network on the back side provides additional thermal paths.
Backside power delivery is one of those technologies that's easy to understate. It sounds like a minor rearrangement of wiring layers. In reality, it's a fundamental change in how chips are manufactured and represents one of the biggest process architecture shifts since the introduction of FinFET transistors.