Once front-end transistors, MOL contacts, and frontside BEOL layers are formed and bonded to a carrier wafer, BSPDN feeds power through thinned silicon . Moving supply rails to the wafer backside decouples resistive voltage drop from frontside signal routing congestion .
Frontside FSPDN vs. Backside BSPDN Architecture
In a conventional frontside power delivery network (FSPDN), power and ground grids share the same upper back-end-of-line (BEOL) metal stack as global and local signal interconnects . For an interconnect wire of resistivity , length , width , and height , line resistance follows , while lateral coupling capacitance rises as wire spacing shrinks . Keeping metal height large while scaling lateral width increases aspect ratios and exacerbates capacitance and fill constraints, whereas reserving wide, thick upper-level tracks for reverse-scaled power distribution consumes routing capacity needed by signals . Furthermore, supply current entering from upper BEOL tiers must traverse a tall chain of resistive vias and contacts down to the middle-of-line (MOL) source/drain regions, compounding resistive drop along the vertical stack .
Backside power delivery networks (BSPDNs) resolve this resource contention by physically separating power distribution from signal routing across opposite faces of the semiconductor device layer . Signal traces remain on the frontside BEOL, where removing power rails from lower metal tiers such as M0 relaxes metal pitch and routing congestion . Simultaneously, wide power and ground rails—such as backside power rails (BPR)—are fabricated beneath the thinned substrate and routed upward through short backside vias or via-to-backside-power-rail (VBPR) structures directly to transistor source/drain nodes and backside gate tie-downs . As summarized in the accompanying concept schematic, separating frontside signal routing from backside power rails breaks the coupling between wire resistance scaling and signal-track crowding .
Process map
40nm BSI CMOS Image Sensor
Use the 40nm BSI CMOS Image Sensor overview to trace wafer bonding, backside thinning, and backside contact handoffs; interactive 2.5D courses cover 40nm–7nm nodes rather than sub-5nm BSPDN flows.
Public flow overview
Wafer Bonding, Thinning, and Backside Via Sequence
Building a BSPDN stack requires strict process ordering across frontside device completion, wafer-level carrier bonding, backside silicon removal, and backside metallization . First, transistors—such as nanosheet or FinFET channels—shallow trench isolation (STI), gate-cut dielectric spacers, MOL dielectrics, frontside source/drain contacts, and frontside BEOL interconnect levels are fabricated on the primary device wafer . At the MOL stage, power-receiving landing structures can be prepared in advance: for example, merging a frontside source/drain contact opening with a VBPR opening defined inside a gate-cut dielectric spacer above the STI creates an enlarged landing interface , while downward-extending gate footings separated from channels by bottom dielectric isolation (BDI) provide dedicated backside landing targets for gate tie-downs .
Second, the completed frontside BEOL stack is bonded to a mechanical carrier wafer through a dielectric bonding layer—such as a bonding oxide—so the fragile active layer remains mechanically supported during subsequent backside removal . As established in direct bond interconnect and hybrid wafer bonding studies, achieving void-free, mechanically robust wafer-to-wafer or die-to-die attachment requires low dielectric surface roughness, surface activation to promote covalent bond formation at ambient temperature, tight particle control, and post-bond thermal annealing to strengthen interfacial bonding . For schemes involving metallic hybrid interfaces, two-step chemical mechanical planarization (CMP) must tightly control copper recess and dielectric erosion because pad-wafer contact mechanics alter local step profiles .
Third, the bonded wafer pair is inverted (flipped), and the original bulk silicon handle substrate is removed in stages . Analogous to why BSI sensors bond before thinning and two-stage backside silicon CMP, bulk thinning combines coarse mechanical grinding and polishing with selective plasma or chemical etching that lands cleanly on a buried etch-stop or isolation layer, followed by controlled etch-stop removal and device-layer recess to expose the underside of the active region and STI . Finally, a backside interlayer dielectric (ILD) is deposited and planarized, backside via openings (VBPR, nano-TSV, or backside gate vias) are etched to expose the target landing structures, and backside power rails are patterned and filled using adhesion liners and low-resistivity metals such as tungsten, ruthenium, or copper . Because backside etching proceeds from the wafer backside inward toward the device layer, both backside vias and power rails naturally exhibit a tapered profile—wider toward the backside surface and narrower toward the frontside landing interface .
Failure Propagation Chain Across Bonding, Thinning, and Power Delivery
Because BSPDN couples frontside lithography, wafer bonding, extreme silicon thinning, and backside metallization across opposite sides of the device layer, defects propagate sequentially along a four-stage causal chain:
- Bonding warpage, interfacial voids, and front-to-back alignment drift: Coefficient of thermal expansion (CTE) mismatch between thin films and silicon substrates drives biaxial thermal stress and wafer curvature , while surface roughness, contamination particles, or non-uniform
CMPdishing generate unbonded interfacial voids whose lateral footprint can exceed the originating particle size by orders of magnitude . Wafer bonding and inversion also introduce alignment offsets between frontside transistor features and subsequent backside lithography . - Backside thinning non-uniformity and etch-stop loss: Wafer bow, bonding-void compliance, or local topography gradients distort pressure uniformity during backside grinding and polishing . Non-uniform silicon removal risks either incomplete etch-stop clearance—leaving blocked backside via paths—or localized punch-through into the
STI, active channel, orVBPRlanding regions . - Backside contact resistance degradation and thermal path obstruction: When backside alignment shifts or tapered via openings under-land, contact overlap shrinks and current crowds sharply near the contact leading edge over a characteristic transfer length, raising contact versus interconnect resistance . Simultaneously, incomplete metal interface closure at grain-boundary triple points , grain-boundary electron scattering , or thermal-annealing-induced microvoids and cracks driven by Cu-to-Si CTE mismatch inside vertical vias further degrade vertical conductance. In parallel, replacing bulk silicon with backside dielectric isolation and an oxide-bonded carrier wafer impedes vertical heat conduction away from active transistor channels .
- Supply IR drop, electromigration acceleration, and timing degradation: Elevated vertical via and contact resistance amplifies resistive voltage drop () across standard cells and power-gating header switches , while higher local current density and thermal stress accelerate vacancy migration and electromigration voiding . The resulting supply droop reduces effective transistor gate overdrive and degrades circuit switching speed and timing margin .
Architecture Variants and Thermal-Mechanical Tradeoffs
Process architects balance contact resistance, standard-cell density, and thermomechanical reliability across distinct BSPDN connection topologies . A merged frontside source/drain contact and VBPR structure enlarges the conductive overlap between the backside rail via and the source/drain contact within a gate-cut spacer boundary, reducing contact sensitivity to front-to-back overlay shift . For transistor isolation, routing a backside gate tie-down via directly from a backside power rail onto a downward gate footing separated by BDI provides an electrical single diffusion break without consuming frontside M0 routing or risking frontside etch damage adjacent to channels . Similarly, in SRAM and logic power-gating header circuitry, multiple dedicated backside power rails (VDD, VSS, and gated virtual rails) sit beneath continuous diffusion regions and feed transistors through vertical vias, eliminating power tracks from M0 altogether .
Every backside vertical interconnect introduces a structural tradeoff between electrical conductivity and thermomechanical stress . Copper offers low bulk resistivity and low delay, yet its large CTE mismatch against surrounding silicon concentrates anisotropic tensile and shear stresses at grain-boundary intersections during post-fill metallization and wafer-bonding thermal cycles, driving microvoids, cracks, or extrusion unless grain microstructure and coherent twin boundaries stabilize grain growth . High-melting refractory or noble liners and fills such as tungsten or ruthenium with titanium-nitride adhesion liners mitigate electromigration and thermal extrusion inside narrow tapered backside vias , while TSV integration timing and thermal budget ordering determine whether vertical vias endure subsequent frontside or bonding anneals .
Connecting Backside Integration to Interactive 40nm–7nm Flows
SemiFlows' interactive 2.5D cross-section courses cover 40nm–7nm planar and FinFET production nodes rather than sub-5nm or 28nm BSPDN flows. Within that verified curriculum, the free 40nm BSI CMOS Image Sensor overview serves as the closest structural bridge for understanding BSPDN process integration: both architectures complete frontside devices and multilevel interconnects first, bond the active wafer face to a supporting wafer, flip and thin the bulk silicon from the backside, and form backside dielectric interfaces and vertical electrical contacts . Walking through the 40nm BSI bonding, two-stage thinning, and backside contact sequence clarifies the mechanical and interface dependencies behind BSPDN without extrapolating unverified fabrication settings or tool parameters .
References
Via-to-backside power rail structure
Zhou Huimei, RAMACHANDRAN RAVIKUMAR, ZHUANG LEI, XIE RUILONG, YANG XIAOMING
US-2025318238-A1 · IBM · Filed 2024
Backside gate tie-down in backside power distribution network (bspdn) architecture
ZHANG CHEN, XIE RUILONG, SARKAR DEBARGHYA, LI TAO
US-2025309116-A1 · IBM · Filed 2024
Header layout design including backside power rail
MORI HARUKI, TIEN CHIEN-CHI, HUANG CHIA-EN, FUJIWARA HIDEHIRO, CHEN YEN-HUEI
US-2023410851-A1 · TAIWAN SEMICONDUCTOR MFG CO LTD · Filed 2023
Microstructure Evolution and Defect Formation in Cu Through-Silicon Vias (TSVs) During Thermal Annealing
Hae-A-Seul Shin, Byoung-Joon Kim, Ju-Heon Kim, Sung-Hwan Hwang, A. Budiman, H.-Y Son et al. · Journal of Electronic Materials
Sub-10µm Pitch Hybrid Direct Bond Interconnect Development for Die-to-Die Hybridization
John P. Mudrick, Jonatan A. Sierra-Suarez, M. Jordan, T. Friedmann, R. Jarecki, M. Henry · Electronic Components and Technology Conference
Silicon VLSI Technology - Full
James D. Plummer, Michael D. Deal, Peter B. Griffin
Silicon VLSI Technology · ISBN 978-0130850379