⚛️ Module 1 · Physics Foundation · Chapter 1.4 · 14 min read

MOSFET — The Atom of 28 nm CMOS

One gate, one channel, all of modern computing.

Prerequisites

What you'll learn here

  • Draw the cross-section of an n-channel MOSFET and name source / drain / gate
  • Explain the threshold voltage V_th and why an inversion channel forms
  • Distinguish triode from saturation (pinch-off); apply the I_D equations
  • Argue why NMOS + PMOS makes a CMOS inverter
  • Say why MOSFETs are the foundation of SIDRACHIP's 28 nm CMOS base die

Hook: 10¹⁸ Switches a Second

Right now, in the device you’re reading this on, roughly 10¹⁸ MOSFETs are flipping every second. Each is about 28 nanometers wide — a quarter of a virus. Each is a controlled valve that turns “gate voltage” into “current”. By the end of this chapter you’ll be able to draw one.

Chapter 1.3 gave us the diode — one-way but uncontrolled. MOSFET is the same idea with a third terminal: the gate. Put a small voltage on the gate, open or close the main current path. SIDRACHIP’s 28 nm CMOS base die has billions of MOSFETs driving the memristor layers above — this chapter is how they work.

Intuition: Does the Channel Open?

Cross-section of an n-channel MOSFET:

  • p-type substrate (bottom, broad) — hole-heavy silicon.
  • n⁺ source and n⁺ drain — two heavily-doped electron pools buried in the substrate.
  • Gate (top, metal or polysilicon) — insulated from substrate by a thin oxide (SiO₂).

With no gate voltage, there’s no channel. Source and drain look like two back-to-back diodes; no current.

Apply a positive voltage to the gate. Electrons in the p-substrate are electrostatically pulled toward the surface under the gate. Past a threshold VthV_{th} (typically 0.3-0.5 V) an inversion layer forms right at the surface: despite being p-substrate, electrons dominate there. That layer connects source to drain — the channel opened.

Now put a positive VDSV_{DS} on the drain and electrons flow source → drain. The channel is a voltage-controlled valve.

Formalism: Square-Law MOSFET

L1 · Intro

Three regimes:

  • Off (VGS<VthV_{GS} < V_{th}): no channel, ID0I_D \approx 0.
  • Triode (VDS<VGSVthV_{DS} < V_{GS} - V_{th}): channel open and uniform; IDI_D grows with both VGSV_{GS} and VDSV_{DS} — like a small resistor.
  • Saturation (VDSVGSVthV_{DS} \geq V_{GS} - V_{th}): channel pinches off at the drain end; current depends almost only on VGSV_{GS}.

For digital switching: triode ≈ “on low-R”, saturation ≈ “on constant-current”. Below VthV_{th} = off.

L2 · Full

Square-law MOSFET equations:

  • Triode:
ID=μnCoxWL[(VGSVth)VDSVDS22]I_D = \mu_n C_{ox} \frac{W}{L} \left[ (V_{GS} - V_{th}) V_{DS} - \frac{V_{DS}^2}{2} \right]
  • Saturation:
ID=12μnCoxWL(VGSVth)2I_D = \frac{1}{2} \mu_n C_{ox} \frac{W}{L} (V_{GS} - V_{th})^2

Terms:

  • μn\mu_n: electron mobility (Si ~400 cm²/V·s).
  • Cox=εox/toxC_{ox} = \varepsilon_{ox}/t_{ox}: gate oxide capacitance per area.
  • W/LW/L: channel width / length (designer’s knob — at 28 nm, L=28L = 28 nm).
  • Vov=VGSVthV_{ov} = V_{GS} - V_{th}: “overdrive”.

CMOS inverter: one n-MOS + one p-MOS driven by the same input. When one is on, the other is off. Output: input 1 → output 0, input 0 → output 1. Power only during transitions (static ≈ 0). The foundation of all digital logic.

L3 · Deep

Short-channel effects below L<100L < 100 nm:

  • V_th roll-off: shorter LL → drain field reaches into the source barrier (DIBL) → lower VthV_{th}.
  • Subthreshold slope: S=ln(10)(kT/q)(1+Cdep/Cox)S = \ln(10) (kT/q)(1 + C_{dep}/C_{ox}). Theoretical floor 60 mV/decade.
  • Leakage: tunneling through the gate oxide — solved by high-k dielectrics like HfO₂ (yes, the same material that becomes our memristor!).
  • Velocity saturation: at high field, electron drift velocity saturates; IDI_D turns linear in VovV_{ov} instead of quadratic.

FinFET (14 nm) and GAA (3 nm): planar control fails, so the gate wraps the channel on 3 sides (FinFET) or all 4 (Gate-All-Around nanosheet).

Relevant for SIDRA: the 28 nm N28HPC+ base die uses conventional planar HKMG (High-K Metal Gate) MOSFETs. HfO₂ appears twice: as the gate dielectric and as the memristor above — same material, two roles.

Experiment: Turn the Gate, Watch the Channel

Try:

  1. VGS=0V_{GS} = 0, VDS=0.5V_{DS} = 0.5: no channel. ID=0I_D = 0. “OFF” chip.
  2. Slowly raise VGSV_{GS}. At Vth=0.4V_{th} = 0.4 V the channel line appears. Current still small.
  3. VGS=0.6V_{GS} = 0.6: channel thicker; if VDS<VGSVthV_{DS} < V_{GS} - V_{th} → “TRIODE”.
  4. Raise VDSV_{DS} — when VDSVGSVthV_{DS} \geq V_{GS} - V_{th} the drain end pinches off, “SATURATION” appears.
  5. VGS=1.2V_{GS} = 1.2, VDS=1.2V_{DS} = 1.2: fully saturated. Current (VGSVth)2\propto (V_{GS} - V_{th})^2.

Channel thickness is drawn proportional to (VGSVth)(V_{GS} - V_{th}); in saturation the drain-end narrowing is visible.

Quiz

1/5How does the channel form in an n-channel MOSFET?

Lab Task: I_D Calculation

Data: μnCox=200\mu_n C_{ox} = 200 µA/V², W/L=10W/L = 10, Vth=0.4V_{th} = 0.4 V.

(a) VGS=1.0V_{GS} = 1.0 V, VDS=1.0V_{DS} = 1.0 V → IDI_D? (Hint: saturation?) (b) VGS=1.0V_{GS} = 1.0 V, VDS=0.3V_{DS} = 0.3 V → IDI_D? (Hint: triode?) (c) VGS=0.3V_{GS} = 0.3 V, VDS=1.0V_{DS} = 1.0 V → IDI_D?

Answers

(a) Vov=0.6V_{ov} = 0.6, VDS(1.0)>Vov(0.6)V_{DS} (1.0) > V_{ov} (0.6) → saturation. ID=12(20010)(0.6)2=10000.36=360I_D = \tfrac{1}{2}(200 \cdot 10)(0.6)^2 = 1000 \cdot 0.36 = 360 µA.

(b) VDS(0.3)<Vov(0.6)V_{DS} (0.3) < V_{ov} (0.6) → triode. ID=(20010)[0.60.30.32/2]=2000[0.180.045]=20000.135=270I_D = (200 \cdot 10)[0.6 \cdot 0.3 − 0.3²/2] = 2000 \cdot [0.18 − 0.045] = 2000 \cdot 0.135 = 270 µA.

(c) VGS(0.3)<Vth(0.4)V_{GS} (0.3) < V_{th} (0.4) → off. ID0I_D ≈ 0 (real subthreshold leakage ignored here).

Cheat Sheet

  • MOSFET: 3-terminal — source (S), drain (D), gate (G). n-channel: S/D are n⁺, substrate is p.
  • Threshold VthV_{th} ≈ 0.3-0.5 V in modern CMOS. V_GS > V_th → inversion channel.
  • Triode: V_DS < V_ov, linear+quadratic; digital 0.
  • Saturation: V_DS ≥ V_ov, pinch-off, IDVov2I_D \propto V_{ov}^2; digital 1.
  • CMOS: NMOS + PMOS = inverter; static power ≈ 0, dynamic = C·V²·f.
  • SIDRACHIP: 28 nm N28HPC+ HKMG MOSFETs under the memristor stack.

Vision: Beyond Planar MOSFET

Planar MOSFET hit a physics wall at 28 nm. Next generations:

  • FinFET (14-7 nm): gate wraps the channel on 3 sides (Intel from 22 nm, TSMC 16 nm). Candidate for SIDRA Y10 evaluation.
  • Gate-All-Around (GAA) nanosheet (3-2 nm): gate wraps all 4 sides. Samsung 3 nm (2022), TSMC N2 (2025), Intel 18A.
  • CFET (Complementary FET): stacked NMOS + PMOS vertically; ~0.5× area.
  • 2D MoS₂ transistor: sub-2 nm channel — IBM 2024 demo, density leap.
  • Neuromorphic synaptic transistor: gate capacitance carries an analog weight — 3-terminal memristor; alternate topology to SIDRA.
  • Spintronic logic: MTJ-based gate — infinite endurance, ferromagnetic-state logic.
  • TFET (Tunnel FET): band-to-band tunneling breaks the 60 mV/dec floor; ultra-low-power IoT.
  • Ferroelectric FET (FeFET): HZO gate → non-volatile logic + memory in one device.
  • Cryogenic CMOS: transistors that operate at 4 K; quantum control chips, superconducting AI hybrids.

Biggest lever for post-Y10 SIDRA: CFET plus a BEOL analog memristor layer — 2× digital-control density from CFET, plus vertically stacked memristors for MVM. Total density 4–5× vs. Y10. 2029–2031 horizon.

Further Reading