← Back to list
Hardware & Semiconductor
#BSPDN#PowerVia#매립전력레일#나노TSV#IR드롭
Last updated · 2026-11-09

Backside Power Delivery Network (BSPDN) and Next-Generation Power-Wiring Innovation

1. Overview

A. Definition

A Backside Power Delivery Network (BSPDN) is a next-generation interconnect technology that moves the power wiring (VDD·VSS) that supplies transistors from the conventional front side of the wafer to the back side of the wafer, physically separating signal wiring from power wiring. With the transistor in the middle, the key idea is to place signal lines on top and power lines underneath, raising power-delivery efficiency and wiring density at the same time.

As semiconductor process nodes enter the 2 nm·1.x nm class (the angstrom era), the bottleneck that throttles performance is no longer the transistor (device) itself but the dozens of layers of wiring (BEOL, Back-End-Of-Line) that cover it. As line widths shrink, wiring has become thinner and denser, and high-speed signals and power now compete for the same space within a single thin metal layer. Thin wiring has high resistance that causes voltage to drop—IR-Drop—and the space occupied by power lines blocks signal routing, creating Routing Congestion. BSPDN tackles both problems head-on with the idea of "sending power lines all the way to the back side so signals and power take separate paths," and together with GAA transistors it is regarded as another structural turning point for leading-edge angstrom-era processes.

B. Background and Necessity

Three pressures converged to make BSPDN inevitable. First is the deepening Power Delivery Bottleneck. In the conventional structure, external power travels from package bumps → the top thick front-side metal → dozens of mid/lower metal layers → to the transistors, passing through wiring that grows ever thinner. When the resistance of this long path accumulates, the voltage actually reaching the transistor falls below its nominal value—an IR drop—which translates directly into lower operating frequency and eroded timing margins. As metal cross-sections shrink with scaling, resistance rises sharply, and below 7 nm this problem emerged as a performance limit.

Second is the spatial competition between signal and power wiring. On the front-side wiring layers, signal lines and power lines must share limited metal resources, and the area the power grid occupies reduces the signal-routing resources available. Designers space cells loosely to avoid congestion, which leads to wasted standard-cell area and lower integration density. Moving power lines to the back side lets the front side be devoted entirely to signals, packing more logic into the same area.

Third is synergy with the GAA transition. [[gaa-transistor]] improved device control, but no matter how good the device is, it cannot deliver performance if power is not supplied properly. To squeeze out PPA (performance·power·area) to the limit in the angstrom era, device innovation and wiring innovation must advance together. As these three pressures converged, the BSPDN concept that imec concretized around 2019 with the Buried Power Rail (BPR) and nano-TSV was lifted from the research stage to volume production. Intel began the world's first high-volume production on its 18A process in 2024–2025 under the name PowerVia, with TSMC (A16, Super Power Rail) and Samsung (SF2 family) following.

C. Key Characteristics

The nature of BSPDN can be summarized in four points.

  • Plane Separation: placing signals (front) and power (back) on different planes to eliminate spatial competition.
  • Low-resistance power path: laying a thick power grid on the back side to greatly lower IR drop and via resistance.
  • Higher integration density: emptying the front side for signals only, so standard cells can be packed more tightly (room for 15–20% area).
  • Combination with device innovation: pairing with [[gaa-transistor]] to optimize angstrom-era PPA to the limit.

2. Structural Difference Between Conventional Power Delivery and BSPDN

The shift in power-delivery method is summarized by "is the power line above or below the transistor." The structure diagram below shows the cross-sectional difference between frontside and backside delivery.

flowchart TB
  subgraph FS["Conventional Frontside PDN"]
    direction TB
    P1["Package bumps (power·signal mixed)"] --> M1["Front-side metal: signal+power compete"]
    M1 --> TR1["Transistor (device)"]
    M1 -. "thin wiring -> IR drop·congestion" .-> M1
  end
  subgraph BS["Backside PDN"]
    direction TB
    S2["Front-side metal: signal only"] --> TR2["Transistor (device)"]
    TR2 --> N2["nano-TSV (nTSV)"]
    N2 --> B2["Back-side power layer: thick low-R power grid"]
    B2 --> P2["Back-side bumps (power only)"]
  end
  FS ==> BS

A. Limits of Conventional Frontside Power Delivery

In the conventional structure the transistor is built on the wafer front side, and above it 10–15+ BEOL metal layers carrying both signal and power are stacked. Power is received at the topmost thick metal and descends to the device through ever-thinner lower layers. The problem is that this path is long and high-resistance, and shares layers with signal lines.

The farther power travels through thin wiring, the more voltage drop accumulates in proportion to resistance, so IR drop worsens in regions of high current demand. According to imec's modeling, via resistance alone is on the order of several hundred Ω (about 300 Ω) for a narrow front-side via pillar, whereas a thick back-side TSV can be lowered to single-digit Ω (about 5 Ω). Moreover, because the power grid occupies a large portion of the front-side metal, signal-routing resources become scarce, producing design congestion and area inefficiency together. In short, frontside delivery is a structure that loses on both power quality (IR drop) and wiring density (congestion) at once.

IR drop matters because it draws the upper bound of performance and reliability. Transistors switch more slowly as the supply voltage falls, so if the power grid's voltage drop is large, designers must raise the nominal voltage to compensate, which in turn increases power consumption and heat—a vicious cycle. In addition, the voltage swing from sudden current changes (di/dt noise) causes timing errors, ultimately acting as a "Power Wall" that conservatively holds the operating frequency low. However much a fine process raises the device's own performance, that performance is unusable if power is not delivered cleanly, which is why power delivery became the real bottleneck of the angstrom era.

B. Backside Structure: Physical Separation of Power and Signal

BSPDN places only signal wiring on the front side and only power wiring on the back side, with the transistor in the middle. Because the two wirings use different planes, spatial competition disappears, and on the back side, with no need to contend with signal lines, a much thicker, lower-resistance power grid can be laid. Thick power lines have low resistance that greatly reduces IR drop, and because power is supplied over a short path right beneath the transistor, delivery efficiency rises.

At the same time, the front-side metal can use the space vacated by power lines entirely for signals, easing routing congestion and letting standard cells be packed more tightly. This is a rare "win-win" structure that improves performance (IR drop↓)·area (density↑)·power efficiency all at once. imec reported that a BPR+backside-delivery combination can improve IR drop by about 7× over conventional frontside delivery, and Intel stated it measured about a 6% frequency gain along with improved cell utilization when applying PowerVia.

C. Key Components: BPR·nano-TSV·Back-side Metal·Wafer Thinning

BSPDN is realized by combining several key process elements. Each element shares one goal: "how to connect power from the back side to the device short and thick."

First, the Buried Power Rail (BPR) is a structure that buries the power rail beneath the transistor (or beside the device region). Because the power rail descends from the front-side metal into the bottom of the cell, it reduces the number of standard-cell tracks, creating room for about 15–20% area scaling. BPR can be combined with frontside delivery, but its IR-drop improvement is maximized when combined with backside delivery.

Second, the nano-TSV (nTSV, nano Through-Silicon Via) is a fine vertical via that pierces the thinned wafer to connect the back-side power grid with the front-side (or BPR) power nodes. imec reports that it landed nTSVs precisely on BPRs at a tight pitch of about 200 nm without consuming standard-cell area. Alignment precision is the crux, requiring front-to-back overlay registration on the order of about 10 nm. Third, the Back-side Metal forms a thick, low-resistance power grid that does not compete with signals, and when needed a back-side capacitor (MIM-CAP) is added to reinforce power stability. Fourth, Wafer Thinning grinds the wafer down to the hundreds-of-nm level after device fabrication to enable back-side processing; it is a prerequisite process performed together with carrier-wafer bonding.

Component Role Key effect Technical challenge
BPR (buried power rail) bury power rail under cell shrink cell tracks (area 15–20%↓) buried-metal contamination·thermal budget
nano-TSV (nTSV) vertical back↔front power link low-R power path (about 5 Ω) 10 nm-class alignment precision
back-side metal grid thick low-R power wiring large IR-drop reduction back-side process heat·stress
wafer thinning·bonding prerequisite for back-side process secures back-side access ultra-thin handling·yield

As the table shows, the value of BSPDN comes not from individual elements but from the combination of "bury + pierce + back-side wiring." Reconstructing the power path to be short and thick lowers IR drop, while simultaneously emptying the front side raises integration—two effects achieved at once in one structure.

3. Manufacturing Process and Divergence in Implementation

BSPDN adds the step "flip the wafer and process the back side" to the existing process. The flow below shows a representative manufacturing sequence.

flowchart LR
  A["Front-side device·BEOL (FEOL+signal wiring)"] --> B["Carrier-wafer bonding"]
  B --> C["Wafer flip"]
  C --> D["Back-side ultra-thinning (hundreds of nm)"]
  D --> E["nano-TSV etch·fill"]
  E --> F["Back-side power metal·bumps"]
  F --> G["Back-side PDN complete·inspection"]
  E -. "10 nm-class front-back alignment" .-> E
  D -. "thin-film stress·warp control" .-> D

A. Typical Manufacturing Flow

First, the transistor (FEOL) and signal-centric front-side wiring are formed on the front side. Next, a carrier wafer is bonded to protect and support the processed front side, and the whole is flipped. Then the originally thick silicon substrate is ultra-thinned from the back to the hundreds-of-nm level, a step in which managing the wafer's warp and stress is crucial. After thinning, nano-TSVs are etched and filled to vertically join the back side with the device-region power nodes, and above them the back-side power metal layer and bumps are formed. Finally, the resistance and connectivity of the back-side PDN are inspected. Compared with the existing process, this adds an entirely new process family—bonding·flip·thinning·back-side lithography—so the burden on equipment, yield, and inspection is large.

The particularly tricky steps in this flow are ultra-thinning and front-back alignment. A wafer hundreds of nm thick warps as easily as a sheet of paper, so even a slight wobble in polishing uniformity or bonding quality can distort device characteristics or cause cracks through stress. Also, when drilling nano-TSVs from the back, they must be matched within about 10 nm to the power nodes already made on the front, making precise overlay metrology that sees through both sides essential. The back-side process's thermal budget is also a constraint: because the front-side device is already complete, back-side processing must stay within a low-temperature process window so that high-temperature steps do not damage the front-side device. Thus the difficulty of BSPDN stems less from individual elements than from the constraint of the process order itself—processing from behind without harming a finished device.

B. Divergence in Implementation: The Connection Point of the Power Tap

BSPDN diverges by generation and vendor according to "where the back-side power connects to the device." imec presented supplying power via the BPR as the baseline, which is favorable for cell-area reduction. Intel's PowerVia uses a scheme in which nano-TSVs land on the cell's internal power taps, focusing on simplifying design and debugging by separating the signal BEOL from the power PDN.

TSMC's Super Power Rail (SPR) goes a step further, choosing to connect back-side power directly to the source·drain terminals (Direct Backside Contact) to further lower contact resistance. This direct-contact approach may have higher power-delivery efficiency but the highest process difficulty. According to reports, TSMC verified SPR on A16 without greatly changing existing design rules, but the volume-production timing is reported variously as late 2026 to 2027, so assertions are avoided. In short, as the connection point deepens from via BPR → cell-internal tap → direct source·drain, power efficiency improves but manufacturing difficulty rises together—a trade-off exists.

4. Comparing Effects and Limits

The advantages of BSPDN are clear, but a balanced judgment requires viewing together the cost the new process adds and the critical limit of the heat-dissipation problem.

Perspective Conventional frontside PDN Backside PDN (BSPDN) Cause of the difference
IR drop large (thin-wiring accumulation) small (up to ~7× improvement) thick low-R back grid·short path
wiring congestion signal·power compete front side for signals only power·signal plane separation
integration density low high (room for ~15–20%) BPR·cell-track reduction
heat dissipation favorable (dissipates to front) unfavorable (device sandwiched) power layer·substrate block heat path
process cost·yield mature high (bonding·thinning·back-side process) new process family added

The key the comparison shows is that BSPDN gives clear gains in performance and area while paying heat and cost as the price. In particular, heat is a structural homework. The conventional structure exposed power devices on the front side, making it easy to extract heat with a heat sink, but BSPDN sandwiches the transistor between the signal layer and the power layer, and the back side where power flows partly blocks the main dissipation path, raising the risk of localized overheating (Hot Spots). The higher the performance and power of the chip, the more prominent this problem, so back-side heat-transfer design and package thermal strategy are cited as variables that decide BSPDN's success.

As a concrete example, Intel created the world's first volume-production case by applying this to Panther Lake (client)·Clearwater Forest (server)-class products combining RibbonFET (GAA) and PowerVia on 18A. imec·ARM demonstrated the IR-drop improvement and area gain of backside delivery with a joint test chip, and showed that adding a back-side MIM-CAP can raise capacitance density 4–5× to further improve power stability. Thus the industrial rollout proceeds rapidly from research (imec) → production (Intel) → followers (TSMC·Samsung).

Practical measures discussed to mitigate the heat-dissipation limit include the following.

  • Using back-side metal for heat transfer: using the thick back-side power metal not only for power but also as a path to spread heat, dispersing localized overheating.
  • Thermal-aware Placement: distributing high-heat blocks and lowering current-density peaks to suppress hot-spot formation.
  • Package·cooling co-design: optimizing package cooling structures such as direct liquid cooling and heat spreaders together, on the premise of back-side dissipation.
  • Adaptive voltage·frequency control ([[dvfs]]): dynamically adjusting voltage·frequency to the thermal state to automatically throttle power in overheated zones.

These measures all share the common goal of "not giving back in heat the gains obtained in power delivery." Where BSPDN sets the balance point among performance·area and heat is what decides a product's competitiveness.

5. Deep Dive: The Angstrom-Era Roadmap and Changes in the Design Ecosystem

BSPDN should be understood not as a single process technology but as an axis that reshapes the entire scaling strategy of the angstrom era. First, from a roadmap perspective, the leading three push GAA (device) and BSPDN (wiring) as a set. Intel secured the narrative of "wiring-innovation leader" by producing 18A (RibbonFET+PowerVia) ahead of others, while TSMC chose phased introduction from N2 (GAA only) → A16 (GAA+SPR) to distribute yield risk. Samsung is also preparing backside delivery in the SF2 family. This shows that "who first mass-produces BSPDN at stable yield" is the watershed of future foundry leadership competition.

Second, the change in the design (EDA)·verification ecosystem is large. With power and signal placed on different sides, the existing 2D·single-side-centric place-and-route (P&R) and power-integrity (IR/EM) analysis flows must be redefined on a double-side basis. New design·verification tools and sign-off criteria are required that also account for the back-side PDN's heat·stress, front-back alignment error, and even back-side capacitors. Third, the elevation of thermal design's importance. As seen earlier, BSPDN is unfavorable for dissipation, so thermal-aware co-design linked with package·cooling—using back-side metal as a heat-transfer path—becomes essential. Fourth, combination with chiplets·3D stacking. When BSPDN meets 3D integration such as [[chiplet]]·[[high-bandwidth-memory]]·CFET, power delivery·dissipation·bonding grow more complex, but at the same time there is great potential to fundamentally improve the power integrity of heterogeneous integrated systems. In short, BSPDN is positioning itself as a system-level innovation that cuts across device·wiring·package·design tools.

6. Considerations and Implications

From a professional engineer's perspective, the following should be weighed in a balanced way when adopting and evaluating BSPDN.

A. Adoption strategy — phased introduction and yield-risk management. Because BSPDN adds an entire new process family—bonding·ultra-thinning·back-side lithography—securing early yield is the biggest gateway. A strategy like TSMC's of not cramming device innovation (GAA) and wiring innovation (BSPDN) into one generation but introducing them separately is a wise approach that distributes risk. Organizations newly adopting it should selectively apply it first to products that desperately need high power·high integration, and run a roadmap of coexistence with the mature front-side process in parallel.

B. Trade-off — performance·area gains vs. heat·cost. BSPDN gives clear gains in IR drop·integration density but at the cost of worsened heat and higher process cost. Especially in chips with extreme power density such as data-center·AI accelerators, dissipation decides the performance ceiling, so without thermal-power co-optimization encompassing back-side heat transfer·package cooling·power-density placement, the gains are hard to realize. Deciding adoption by performance figures alone will be paid back at the thermal bottleneck.

C. Outlook — foundry competition and standardization. The ability to mass-produce BSPDN stably is a key variable dividing angstrom-era foundry leadership. Intel's preemptive production, TSMC's direct contact (SPR), and Samsung's pursuit enter full competition around 2026–2027, and de facto standardization of design rules·EDA sign-off·inspection methods is expected to follow. Note, however, that concrete figures such as production timing and yield differ across sources, so cross-check vendor official announcements in decision-making and avoid definitive assumptions.

D. Related technologies — integration with GAA·chiplets·advanced packaging. BSPDN composes angstrom systems together with [[gaa-transistor]] (device), [[chiplet]]·[[ucie-chiplet-interconnect]] (heterogeneous integration), and [[high-bandwidth-memory]] (memory stacking). Because power delivery·dissipation·bonding·alignment are intertwined, one must go beyond individual-technology optimization to system-level co-design. In particular, power integrity·thermal management when combined with 3D stacking will be the practical bottleneck of next-generation high-performance computing.

E. Workforce·supply-chain perspective. The back-side process demands new equipment·material·metrology capabilities such as wafer bonding·thinning·back-side inspection, so organizations must preemptively prepare related process·design·verification talent and secure equipment·material supply chains to avoid falling behind in competition.

References


In one line: BSPDN is an angstrom-era wiring innovation that moves power wiring to the back side of the wafer to separate the paths of signal and power, lowering IR drop and raising integration density; paired with GAA it reshapes leading-edge processes, but it is a technology whose price—heat·yield·cost—must be solved through co-optimization.