Power Semiconductor Fundamentals & Power Diodes
What makes a device a power device. Learn voltage blocking with lightly doped drift regions, the unipolar silicon limit (R_on,sp ∝ BV^2.5) and how conductivity modulation in bipolar devices beats it, current conduction and on-state loss, switching loss, safe operating area (FBSOA, RBSOA, SCSOA), thermal limits, avalanche ruggedness, and power diodes — PiN power diodes, fast/ultrafast/soft recovery, Schottky and SiC Schottky, reverse recovery and Q_rr, snubbers and rectifier applications — with Python computing drift thickness, doping and specific on-resistance for Si, SiC and GaN at 600 V and 1200 V, the Baliga figure of merit, and reverse-recovery losses.
Power devices must block hundreds or thousands of volts when off and carry tens to thousands of amperes when on, while switching fast and surviving faults. These requirements conflict: a thick, lightly doped drift region blocks voltage but resists current.
Understanding that trade-off — and the tricks that bend it (conductivity modulation, superjunctions, wide-bandgap materials) — explains the whole power device landscape.
An analogy
A power device's drift region is like a dam wall: thicker and stronger (lightly doped) to hold back more water (voltage), but a thick wall also makes it harder to open a wide spillway (low resistance) when you need flow.
Blocking and conduction
Blocking voltage is supported by the depletion region of a reverse-biased junction in the drift layer. Unipolar devices (MOSFETs, Schottky diodes) conduct with majority carriers only, so drift resistance rises steeply with voltage rating. Bipolar devices (PiN diodes, IGBTs, thyristors) flood the drift region with injected electrons and holes (conductivity modulation), lowering its resistance dramatically — at the cost of stored charge and slower switching.
Power diodes
Families and trade-offs:
- Standard PiN rectifiers: high voltage, low V_F, slow (µs Q_rr) — line frequency
- Fast/ultrafast (lifetime-controlled): 20–100 ns t_rr, higher V_F
- Soft-recovery (tailored doping/lifetime profiles): low overshoot and EMI
- Si Schottky: lowest V_F, no Q_rr, ≤ ~200 V
- SiC Schottky/MPS: 650–1700 V, only capacitive charge — PFC and inverters
Safe operating area and ruggedness
Forward-bias SOA (linear operation: current × voltage × time), reverse-bias SOA (turn-off with inductive loads), short-circuit SOA (IGBTs survive ~5–10 µs), avalanche energy (E_AS for MOSFETs with unclamped inductive switching), surge current (I²t for diodes and thyristors). Linear-mode operation of modern trench MOSFETs is limited by thermal instability (hot spots) — use devices specified for linear SOA.
Device selection trade-offs
Compare devices along Device → voltage → current → frequency → losses → control → application. Low-voltage, high-frequency: Si MOSFETs, GaN. Mid-voltage (600–1200 V), high frequency: superjunction MOSFETs, SiC MOSFETs, GaN (≤ 650 V). High power, lower frequency: IGBTs. Very high power/voltage: thyristors, IGCTs, press-pack devices. Cost, availability, gate-drive complexity and reliability are equally part of the decision.
Formulas
- Drift design (triangular field):
W_D = 2BV/E_c; N_D = εE_c²/(2qBV) - Unipolar limit:
R_on,sp = 4BV²/(εµE_c³) - Silicon power law:
R_on,sp ≈ 5.9 × 10⁻⁹·BV^2.5 Ω·cm² (Si, E_c doping-dependent) - Losses:
P = P_cond + (E_on + E_off)·f + Q_rr·V·f + gate drive - Junction temperature:
T_J = T_C + P·θ_JC
Worked example: Drift thickness
Given: Si, 600 V, E_c 3 × 10⁵ V/cm
Find: W_D.
- 2 × 600/3e5
Answer: 40 µm
Worked example: Device resistance
Given: R_on,sp 5 mΩ·cm², die 0.25 cm²
Find: R_on (drift only).
- 5 mΩ·cm²/0.25 cm²
Answer: 20 mΩ
Worked example: Q_rr loss
Given: Q_rr 100 nC, 400 V, 65 kHz
Find: Switch extra loss ≈ Q_rrVf.
- 100e−9 × 400 × 65e3
Answer: 2.6 W
Vertical power device and SOA
Source/anode metal (top)
── p body / p⁺ anode ──
░░░░░ n⁻ drift region: blocks voltage (thickness ∝ BV, doping ∝ 1/BV) ░░░░░
── n⁺ substrate ──
Drain/cathode metal (bottom) current flows vertically through the whole chip area
SOA (log I vs log V): current limit │ R_DS(on) line ╲ thermal limit ╲ second breakdown ╲ BVPower devices are vertical so current uses the full chip area and the drift region can be made as thick as the voltage requires. The safe operating area bounds combinations of voltage, current and time the device survives — including thermal and second-breakdown limits in linear operation.
Drift-region design and material limits
import math
# Unipolar drift-region limit: specific on-resistance vs breakdown voltage for Si, 4H-SiC and GaN
eps0 = 8.854e-14
mats = {"Si": (11.7, 1350, 0.3e6), "4H-SiC": (9.7, 900, 2.5e6), "GaN": (9.0, 1200, 3.3e6)} # eps_r, mu, Ec (V/cm)
for BV in (600, 1200):
for m, (er, mu, Ec) in mats.items():
eps = er * eps0
Wd = 2 * BV / Ec # triangular field
Nd = eps * Ec**2 / (2 * 1.602e-19 * BV)
Ron = 4 * BV**2 / (eps * mu * Ec**3) # ohm*cm^2
print(f"{BV:4d} V {m:6s}: drift {Wd*1e4:6.1f} um, Nd {Nd:.1e} cm^-3, Ron,sp {Ron*1e3:8.3f} mohm*cm^2")
print(f"Baliga figure of merit eps*mu*Ec^3 relative to Si: SiC {9.7*900*2.5e6**3/(11.7*1350*0.3e6**3):.0f}x, GaN {9.0*1200*3.3e6**3/(11.7*1350*0.3e6**3):.0f}x")
# Reverse-recovery loss of a power diode in a 400 V, 100 kHz converter
Qrr, V, f = 300e-9, 400, 100e3
print(f"Qrr 300 nC at 400 V, 100 kHz: diode loss ~ Qrr*V*f/4 = {0.25*Qrr*V*f:.1f} W + extra switch turn-on loss ~ Qrr*V*f = {Qrr*V*f:.1f} W")
# 600 V Si : drift 40.0 um, Nd 4.8e+14 cm^-3, Ron,sp 38.136 mohm*cm^2
# 600 V 4H-SiC: drift 4.8 um, Nd 2.8e+16 cm^-3, Ron,sp 0.119 mohm*cm^2
# 600 V GaN : drift 3.6 um, Nd 4.5e+16 cm^-3, Ron,sp 0.042 mohm*cm^2
# 1200 V Si : drift 80.0 um, Nd 2.4e+14 cm^-3, Ron,sp 152.546 mohm*cm^2
# 1200 V 4H-SiC: drift 9.6 um, Nd 1.4e+16 cm^-3, Ron,sp 0.477 mohm*cm^2
# 1200 V GaN : drift 7.3 um, Nd 2.3e+16 cm^-3, Ron,sp 0.168 mohm*cm^2
# Baliga figure of merit eps*mu*Ec^3 relative to Si: SiC 320x, GaN 910x
# Qrr 300 nC at 400 V, 100 kHz: diode loss ~ Qrr*V*f/4 = 3.0 W + extra switch turn-on loss ~ Qrr*V*f = 12.0 WA 1200 V silicon unipolar device needs an 80 µm drift layer doped only 2.4 × 10¹⁴ cm⁻³, giving ~150 mΩ·cm² — far too resistive, which is why silicon uses bipolar IGBTs above ~600 V. SiC and GaN's ~10× higher critical field lets a 10 µm, 100× more heavily doped layer do the same job, cutting on-resistance by 300–900× (BFOM). Reverse recovery of a 300 nC diode costs ~15 W at 400 V/100 kHz once the switch's extra turn-on loss is included.
Common mistakes
- Wrong: Using a unipolar Si MOSFET at 1200 V for high current. — Right: Silicon's limit makes IGBTs or SiC better.
- Wrong: Ignoring Q_rr in hard-switched bridges. — Right: It adds loss to the switch and EMI.
- Wrong: Operating trench MOSFETs in linear mode without SOA check. — Right: Thermal instability destroys them.
Applications
- Power supplies: PFC and DC–DC stages.
- Motor drives & EV: Inverter switches and diodes.
- Renewables: Solar inverters, wind converters.
- Grid: HVDC and FACTS with thyristors/IGCTs.
Key takeaways
- Thick, lightly doped drift layers block voltage.
- Unipolar R_on,sp ∝ BV²/(εµE_c³).
- Bipolar devices beat the limit via conductivity modulation.
- WBG materials shift the whole trade-off by ~2–3 decades.
Going further
- Superjunction principle: Charge-balanced pillars beat the 1-D limit.
- Edge termination: Guard rings, JTE, field plates.
- Cosmic-ray failure: FIT derating of high-voltage devices.