What Is CMOS Device Architecture

A CMOS device architecture is the foundational physical, electrical, and materials organization of Complementary Metal-Oxide-Semiconductor technology—co-integrating complementary pairs of $N$-channel (NMOS) and $P$-channel (PMOS) field-effect transistors on a shared monocrystalline silicon substrate. Invented by Frank Wanlass and Chih-Tang Sah at Fairchild Semiconductor in 1963, the genius of the CMOS device architecture lies in its push-pull circuit topology: in any static logic state (output high or output low), one transistor is completely switched OFF while the other is switched ON. Because current only flows dynamically during switching transitions to charge or discharge load capacitance ($P_{\text{dyn}} = \alpha C_L V_{\text{DD}}^2 f$), static standby dissipation is virtually zero ($P_{\text{stat}} = I_{\text{leak}} V_{\text{DD}}$), yielding unprecedented energy efficiency that enabled modern multi-billion-transistor microprocessors. Over six decades of scaling, the CMOS device architecture has evolved across four architectural paradigms: (1) *Planar Bulk CMOS* governed by classical Dennard scaling; (2) *Strained Silicon and High-$\kappa$ Metal Gate (HKMG)* to overcome polysilicon gate depletion and direct quantum mechanical gate dielectric tunneling ($I_{\text{gate}} \propto e^{-2\alpha t_{\text{ox}}}$); (3) *3D Tri-Gate FinFETs* providing superior three-sided electrostatic gate wrap-around; and (4) *Gate-All-Around (GAA) Nanosheets (MBCFET)* with vertically stacked horizontal silicon ribbons surrounded on all four sides by the gate electrode, continuing to modern Backside Power Delivery Networks (BSPDN) and Complementary FETs (CFET). At a high level, the architectural anatomy of a CMOS device and its rail-to-rail dynamic switching lifecycle looks like this:

Input logic transition (V_IN)
Rail-to-rail voltage transition ramps from 0V to V_DD or V_DD to 0V
Dual-gate electrostatic inversion
Simultaneously modulates NMOS electron channel and PMOS hole channel
Complementary dynamic commutation cyclePush-pull switching, capacitive load charging/discharging, and zero static leakage
PMOS pull-up conduction (V_IN = 0)
PMOS turns ON; charges load capacitance C_L to V_DD with zero DC path to GND
Dynamic switching transition
Brief crowbar shoot-through current during simultaneous partial V_th conduction
NMOS pull-down conduction (V_IN = V_DD)
NMOS turns ON; discharges C_L rapidly to GND while PMOS is cut off
Static quiescent retention
Zero DC static current; off-state leakage limited to subthreshold I_off < 100pA/μm
↻ Multi-GHz logic switching
Near-zero static standby power
Enables multi-billion-transistor computing within manageable thermal envelopes
Robust rail-to-rail noise margins
Full V_DD-to-GND voltage swing provides unmatched noise immunity across PVT

A few principles are essential for understanding how CMOS device architectures are constructed, how complementary carrier transport operates, and why gate electrostatic confinement evolved from 2D planar to 3D gate-all-around architectures:

Device structural hierarchy and physical classifications. Modern CMOS device architectures are constructed through four generational device technologies:
1. *Planar Twin-Well Bulk CMOS Architecture:*
- *Complementary Well Topology:* Fabricated by defining an $N$-well in a $p$-type silicon wafer to host the PMOS transistor, alongside a $P$-well for the NMOS transistor. Heavy $N^+$ source/drain diffusions form the NMOS conduction terminals, while $P^+$ diffusions form the PMOS terminals.
- *Dielectric Isolation:* Shallow Trench Isolation (STI) trenches filled with $SiO_2$ physically separate adjacent transistors to eliminate parasitic bipolar latch-up (PNPN four-layer thyristor firing).
2. *High-$\kappa$ Metal Gate (HKMG) & Strain Engineering:*
- When gate oxide thickness scaled below $1.2\text{ nm}$ ($~4$ silicon atoms), direct quantum tunneling through $SiO_2$ created unbearable gate leakage currents ($>100\text{ A/cm}^2$).
- The HKMG architecture replaced $SiO_2$ with atomic-layer-deposited hafnium dioxide ($HfO_2$, dielectric constant $\kappa \approx 25$), permitting a physically thicker physical gate dielectric ($t_{\text{phys}} \approx 2\text{–}3\text{ nm}$) while maintaining an equivalent oxide thickness ($\text{EOT} = t_{\text{phys}} \cdot \kappa_{\text{ox}} / \kappa_{\text{high-k}} < 0.8\text{ nm}$). Dual work-function metal gates eliminated polysilicon gate depletion.
- *Strain Engineering:* Uniaxial lattice strain is engineered to boost carrier mobilities: embedding epitaxially grown Silicon-Germanium ($Si_{1-x}Ge_x$) in PMOS source/drain regions induces compressive channel strain, boosting hole mobility by $>100\%$, while Silicon-Carbon ($Si:C$) or tensile silicon nitride capping layers induce tensile strain in NMOS channels to double electron mobility.
3. *3D Tri-Gate FinFET Architecture ($22\text{nm to }3\text{nm}$):*
- As gate length scaled below $25\text{ nm}$, planar CMOS suffered severe Short-Channel Effects (SCE) and Drain-Induced Barrier Lowering (DIBL), where the drain electric field penetrates deep through the substrate to punch through the channel.
- The FinFET architecture erects a thin vertical silicon fin ($W_{\text{fin}} \approx 5\text{–}7\text{ nm}$, height $H_{\text{fin}} \approx 45\text{–}65\text{ nm}$). The gate wraps around three sides of the fin, pinching the channel from both sides and the top, establishing a natural electrostatic scale length:

$$\lambda_{\text{FinFET}} = \sqrt{\frac{\varepsilon_{\text{Si}}}{2\varepsilon_{\text{ox}}} W_{\text{fin}} t_{\text{ox}}} < \frac{L_g}{4}$$

which suppresses subthreshold swing ($SS < 70\text{ mV/dec}$) and off-state leakage.
4. *Gate-All-Around (GAA) Nanosheet & CFET Architecture ($2\text{nm}$ and Beyond):*
- GAA replaces vertical fins with vertically stacked horizontal silicon nanosheets (width $W_{\text{ns}} \approx 20\text{–}50\text{ nm}$, thickness $T_{\text{ns}} \approx 5\text{ nm}$). The high-$\kappa$ gate dielectric and metal gate wrap completely around all four surfaces of each nanosheet, delivering optimal electrostatic gate control.
- *Complementary FET (CFET):* Monolithically stacks the PMOS nanosheets directly on top of the NMOS nanosheets in the third dimension, collapsing standard 6T SRAM cell and standard logic cell areas by $50\%$.

The physical cross-section of a Complementary MOS (CMOS) inverter architecture—contrasting an NMOS transistor and PMOS transistor with High-$\kappa$ Metal Gates and Shallow Trench Isolation—is illustrated below:

Complementary MOS (CMOS) Device Architecture Co-integrated NMOS and PMOS pair with High-κ Metal Gate (HKMG) and STI isolation Cross-Section: Twin-Well CMOS Cell P-Substrate / P-Well (NMOS) N-Well (PMOS) STI N+ N+ Gate_N NMOS e⁻ P+ P+ Gate_P PMOS h⁺ N-Well in P-Substrate | High-κ HfO2 (Purple) | STI (Gray) CMOS Power Dissipation & Electrostatics 1. Total CMOS Power Dissipation: P_tot = α·C_L·V_DD²·f + V_DD·(I_sub + I_gate) Zero static DC current; power consumed only when switching states 2. Subthreshold Swing & Thermal Limit: SS = ln(10)·(k_B·T / q)·(1 + C_dep / C_ox) > 60 mV/dec 3D FinFET and GAA nanosheets drive SS close to ideal 60 mV/dec 3. Noise Margin Immunity (NM_H & NM_L): NM_H = V_OH - V_IH | NM_L = V_IL - V_OL ≈ V_DD / 2 Symmetric β_n = β_p sizing yields unmatched rail-to-rail noise rejection V_DD: 0.7V | EOT < 0.8nm | Ion/Ioff > 10⁶ | Leakage < 10 pA/μm Push-Pull Symmetry Zero DC static quiescent power High-κ Metal Gate HfO2 dielectric suppresses tunneling STI & Well Isolation Eliminates parasitic latch-up CMOS device architecture pairs NMOS and PMOS transistors in complementary push-pull networks delivering near-zero static power consumption and rail-to-rail noise immunity for billion-transistor ICs

In addition to 2D planar and HKMG structures, the modern frontier of CMOS device architecture is defined by Gate-All-Around (GAA) nanosheets and 3D Complementary FET (CFET) stacking. The physical architecture of GAA nanosheets and CFET is illustrated below:

Advanced CMOS Evolution: GAA Nanosheet & 3D CFET Gate electrode completely encapsulates horizontal silicon channels; 3D PMOS-on-NMOS stacking GAA Nanosheet (MBCFET) Silicon Substrate & Bottom Dielectric Isolation Work-Function Metal Gate (Wraps 360°) Si Nanosheet 3 (T = 5nm) Si Nanosheet 2 (T = 5nm) Si Nanosheet 1 (T = 5nm) Variable Sheet Width W_ns: 20nm to 50nm 3D Stacked CFET (PMOS over NMOS) Backside Power Delivery Network (BSPDN) Bottom Transistor: NMOS Nanosheets Tensile Silicon Channel · Source/Drain Epitaxy Middle Dielectric Isolation (MDI Layer) Top Transistor: PMOS Nanosheets Compressive SiGe Channels · Vertical 3D Integration Direct Shared Gate Column Connects P and N Channels CFET Breakthrough: Stacking PMOS directly over NMOS collapses standard logic cell area by 50%, sustaining Moore's Law Gate-All-Around nanosheets wrap gate dielectrics around all channel surfaces to eliminate leakage while 3D CFET vertical stacking doubles transistor density per unit silicon area

Carrier mobility matching, drive current balancing ($\beta_n / \beta_p$), and thermal scaling limits. Designing high-performance CMOS device architectures requires solving fundamental asymmetry between electrons and holes:
- *Effective Channel Width & Mobility Asymmetry:* In silicon, electron bulk mobility ($\mu_n \approx 1400\text{ cm}^2/\text{V}\cdot\text{s}$) is approximately 2.5 to 3 times higher than hole mobility ($\mu_p \approx 450\text{ cm}^2/\text{V}\cdot\text{s}$). In planar CMOS architectures, to achieve equal rise and fall propagation delays ($t_{\text{PLH}} \approx t_{\text{PHL}}$) and symmetric switching thresholds ($V_{\text{inv}} \approx V_{\text{DD}}/2$), circuit designers had to size PMOS channel widths two to three times wider than NMOS ($W_p / W_n \approx 2.5$). In 3D FinFET architectures, where fin width and height are quantized ($W_{\text{eff}} = N_{\text{fin}} \cdot [2H_{\text{fin}} + W_{\text{fin}}]$), matching PMOS drive current required allocating multiple fins (e.g., 2-fin PMOS vs. 1-fin NMOS), increasing layout track heights. With Gate-All-Around nanosheets, channel width $W_{\text{ns}}$ is lithographically adjustable per transistor, allowing continuous fine-tuning of drive current without area waste.
- *Short-Channel Effects (SCE) & Drain-Induced Barrier Lowering (DIBL):* In sub-$20\text{ nm}$ gate lengths, the potential barrier between source and channel is lowered not only by the gate voltage, but also by the drain voltage:

$$\text{DIBL} = \frac{V_{\text{th}}(V_{\text{DD}}) - V_{\text{th}}(V_{\text{DD,low}})}{V_{\text{DD}} - V_{\text{DD,low}}} \quad \left[\frac{\text{mV}}{\text{V}}\right]$$

High DIBL degrades subthreshold swing, causing off-state leakage current ($I_{\text{off}}$) to explode exponentially:

$$I_{\text{off}} = I_0 \cdot 10^{-\frac{V_{\text{th}}}{SS}}$$

GAA architectures, by surrounding the channel with four gate surfaces, reduce the electrostatic scale length to $\lambda_{\text{GAA}} = \sqrt{\frac{\varepsilon_{\text{Si}}}{4\varepsilon_{\text{ox}}} T_{\text{ns}} t_{\text{ox}}}$, clamping DIBL below $30\text{ mV/V}$ and maintaining $I_{\text{on}}/I_{\text{off}}$ ratios exceeding $10^6$.
- *Backside Power Delivery Networks (BSPDN):* In traditional CMOS architectures, signal wires and high-current power grids ($V_{\text{DD}}$ and $V_{\text{SS}}$) compete for routing resources in the Frontside Back-End-of-Line (BEOL) metal stack, causing severe IR voltage drops ($>10\%$) and RC interconnect bottlenecks. Next-generation CMOS architectures move the entire power distribution network to the backside of the silicon wafer using Nano-Through-Silicon Vias (nTSVs) and Buried Power Rails (BPR). This eliminates power-routing congestion on the frontside, lowers supply resistance, and frees up lower metal layers strictly for high-speed signal routing.

Read a CMOS device architecture through a *complementary-electrostatic-scaling* lens: an endlessly evolving push-pull semiconductor framework where symmetric electron and hole conduction, atomic-layer gate dielectric confinement, and vertical 3D nanosheet stacking power the computing engines of the modern world.

Take what is cmos device architecture further

Ask the copilot about this term, or have our engineers assess it against your process.

Ask CFSGPT about this → Request a technical assessment → Open the CFS app