Semiconductor Devices in Analog, Digital, Power & Communication Systems

Connecting device physics to systems. See how device properties become circuit behaviour: in analog circuits (g_m, r_o, matching, noise and temperature in amplifiers, current mirrors, differential pairs, references, op-amps, filters, oscillators, regulators and sensor interfaces), digital circuits (CMOS inverters and gates, noise margins, delay, dynamic and leakage power, fan-out, SRAM/DRAM/flash, processors), power systems (switch, diode and driver selection), and communication systems (LNAs, mixers, PAs, detectors, photodiodes, lasers, PIN switches, varactors, PLLs, HEMTs, HBTs) — with Python linking device parameters to circuit metrics.

Every circuit block inherits its performance from device physics: an amplifier's gain from g_m·r_o, a reference's accuracy from V_BE's temperature dependence, a processor's power from C·V²·f and leakage, a radio's sensitivity from transistor noise.

Seeing this chain — physics → device model → circuit block → system — is what makes a device engineer useful to circuit designers and vice versa.

An analogy

Device physics is the grammar, device models are the vocabulary, circuit blocks are sentences, and systems are stories. You can write stories without studying grammar, but you'll write better ones if you know why the words behave as they do.

Analog circuits

Device properties that matter:

Digital circuits

CMOS inverters and gates rely on complementary devices for rail-to-rail swing and near-zero static current; V_T sets noise margins and leakage; mobility and capacitance set delay; SRAM needs balanced device strengths; DRAM relies on ultra-low-leakage access transistors and high-k capacitors; flash on floating-gate/charge-trap physics. Processor performance per watt tracks device scaling (FinFET/GAA) and voltage.

Power systems

Converters combine switches (MOSFET, IGBT, SiC, GaN), diodes (Schottky, fast recovery, SiC), gate drivers and passives. Device choices set efficiency (conduction + switching), switching frequency (size of magnetics), EMI (dv/dt, Q_rr), thermal design and robustness (avalanche, short circuit).

Communication systems

Receivers: LNAs (HEMT, SiGe HBT, RF-SOI), mixers and detectors (Schottky diodes, FETs), PLL VCOs (varactors, HBTs). Transmitters: PAs (GaN HEMT, GaAs HBT, LDMOS). Switches: PIN diodes, RF-SOI. Optical links: lasers (DFB, VCSEL) and photodiodes (PIN, APD) with TIAs. Device f_T/f_max, noise and linearity define system sensitivity and range.

Formulas

Worked example: g_m at equal current

Given: 1 mA; BJT vs MOS at V_ov 0.2 V

Find: g_m.

Answer: BJT ~3.9× more

Worked example: Stages needed

Given: Target gain 10⁴, intrinsic gain 100 per stage

Find: Stages.

Answer: 2 stages (plus margin)

Worked example: Receiver output

Given: R 0.5 A/W, 2 µW, 50 kΩ

Find: V_out.

Answer: 50 mV

From physics to systems

  Device physics (bands, junctions, transport)
        ↓
  Transistor model (I_S, β, g_m, r_o, C's, noise, V_T, µ)
        ↓
  Circuit block (amplifier, mirror, reference, inverter, SRAM, LNA, switch)
        ↓
  System (sensor front end, processor, power converter, radio, EV inverter)

Each level abstracts the one below. When a system misbehaves (noise, drift, leakage, EMI), the explanation often lies two levels down in device physics — for example thermal noise from channel resistance, leakage from subthreshold conduction, or drift from BTI.

Device parameters to circuit metrics

import math
# Device parameters -> circuit metrics: gm efficiency, intrinsic gain, CMOS power, photoreceiver output
VT = 0.02585
print(f"BJT gm/I = 1/VT = {1/VT:.1f} /V; MOSFET gm/I at Vov 0.2 V = {2/0.2:.1f} /V, at 0.1 V = {2/0.1:.1f} /V")
print(f"intrinsic gain: BJT VA 50 V -> {50/VT:.0f}; MOSFET lambda 0.1/V at Vov 0.2 V -> {2/(0.1*0.2):.0f}")
alpha, C, V, f, Ileak = 0.15, 2e-9, 0.8, 3e9, 0.5
print(f"processor: dynamic {alpha*C*V*V*f:.2f} W + leakage {Ileak*V:.2f} W")
R, P, Rf = 0.9, 10e-6, 20e3
print(f"1550 nm receiver: {R} A/W x {P*1e6:.0f} uW x {Rf/1e3:.0f} kohm TIA = {R*P*Rf*1e3:.0f} mV")
# BJT gm/I = 1/VT = 38.7 /V; MOSFET gm/I at Vov 0.2 V = 10.0 /V, at 0.1 V = 20.0 /V
# intrinsic gain: BJT VA 50 V -> 1934; MOSFET lambda 0.1/V at Vov 0.2 V -> 100
# processor: dynamic 0.58 W + leakage 0.40 W
# 1550 nm receiver: 0.9 A/W x 10 uW x 20 kohm TIA = 180 mV

The BJT's fixed g_m/I of 38.7 V⁻¹ beats a MOSFET in strong inversion (10 V⁻¹ at 0.2 V overdrive) — why BJTs excel in low-noise, high-g_m analog; MOSFETs approach it only in weak inversion. Device intrinsic gain (BJT ~1900, short-channel MOS ~100) sets the number of stages needed. A 2 nF-switched processor at 0.8 V/3 GHz burns ~0.6 W dynamic plus 0.4 W leakage. A 0.9 A/W photodiode and 20 kΩ TIA turn 10 µW into 180 mV.

Common mistakes

Applications

Key takeaways

Going further