Capstones: Characterization Platform & Power Device Analysis

Two professional capstones. (1) Semiconductor device characterization platform: import measurement data, plot I–V/C–V curves, extract parameters (I_S, n, R_s, V_T, µ, β, V_A, doping), compare temperatures, fit curves, generate SPICE models, compare simulation with measurement, compute errors, compare with datasheets, classify devices and export engineering reports, with explanations of every parameter. (2) Power semiconductor analysis: compare Si MOSFET, IGBT, SiC and GaN from datasheet parameters under stated operating conditions — conduction and switching losses, thermal model and junction temperature, efficiency, and a design recommendation based on requirements rather than universal ranking — with Python comparing four device technologies at 20, 100 and 300 kHz.

The capstones combine everything in the course: physics to interpret curves, measurement to obtain them, modelling to capture them, and engineering judgement to select devices.

Both produce professional deliverables — a working tool and an engineering report — suitable for a portfolio.

An analogy

Capstone 1 is building a doctor's diagnostic lab for devices (measure, analyse, diagnose, report); capstone 2 is being the doctor who prescribes the right device for a patient's specific condition.

Capstone 1 requirements

Deliverable capabilities:

Capstone 2 requirements

Use datasheet parameters (R_DS(on)(T), V_CE(sat), E_on/E_off vs current and temperature, Q_rr, θ values) for MOSFET, IGBT, SiC and GaN candidates; define operating conditions (voltage, current, frequency, temperature, cooling); compute conduction and switching losses iterated with junction temperature; build a thermal model; compute efficiency; and give a recommendation tied to the stated requirements and constraints (cost, EMI, gate drive, availability).

Engineering report

Structure: objective and requirements, device/sample information, method and equipment (with uncertainties), results (plots, tables), extraction/analysis with assumptions, model validation, comparison with datasheets, risks and limitations, recommendation, appendices (raw data, scripts). Traceability from each conclusion to data is the mark of professional work.

Stretch goals

Automate instruments with PyVISA, add C–V-based doping profiles, add pulsed I–V to separate self-heating, include Monte Carlo of device spread in the power analysis, model thermal transients with Foster networks, and add reliability estimates (power-cycling life) to the device comparison.

Formulas

Worked example: RMSE

Given: Errors 2, −1, 1, −2 mV

Find: RMSE.

Answer: 1.58 mV

Worked example: Loss at 100 kHz

Given: SiC: 6.75 W cond + 60 µJ × 100 kHz

Find: P.

Answer: 12.75 W

Worked example: Efficiency

Given: 6 kW, 4 devices × 12.75 W

Find: η (switches only).

Answer: 99.2%

Capstone 1: characterization platform pipeline

  Device ─► measurement data (CSV from SMU/curve tracer, multi-temperature)
        ─► I–V / C–V plots (linear, semilog, Gummel, 1/C²)
        ─► parameter extraction (fits with confidence intervals)
        ─► device model (.model / subckt) ─► SPICE simulation (ngspice batch)
        ─► model vs measurement overlay + error metrics
        ─► datasheet comparison (min/typ/max) ─► device classification (pass/fail/bin)
        ─► engineering report export (PDF/HTML) + saved project + checkpoint quizzes

Each stage produces an artefact that the next consumes. Explanations of every extracted parameter (what it means physically, which region it was extracted from, and its uncertainty) are part of the deliverable, not an afterthought.

Capstone 2: four technologies across frequency

import math
# Capstone: compare MOSFET / IGBT / SiC / GaN for a 400 V, 15 A rms, hard-switched stage at 3 frequencies
devs = {       # conduction model, Esw (uJ) at 400 V/15 A, theta_jc+cs+sa (C/W)
    "Si SJ MOSFET":  (lambda I: I * I * 0.080, 120, 1.2),
    "Si IGBT":       (lambda I: 0.9 * I * 0.9 + 0.03 * I * I, 400, 1.0),
    "SiC MOSFET":    (lambda I: I * I * 0.060, 60, 1.2),
    "GaN HEMT":      (lambda I: I * I * 0.070 * 1.2, 25, 1.8),        # includes dynamic Rds(on) x1.2
}
I, Ta = 15.0, 50.0
for f in (20e3, 100e3, 300e3):
    row = []
    for name, (cond, Esw, th) in devs.items():
        P = cond(I) * 0.5 + Esw * 1e-6 * f
        row.append(f"{name}: {P:5.1f} W/Tj {Ta + P*th:5.0f} C")
    print(f"{f/1e3:3.0f} kHz | " + " | ".join(row))
#  20 kHz | Si SJ MOSFET:  11.4 W/Tj    64 C | Si IGBT:  17.4 W/Tj    67 C | SiC MOSFET:   8.0 W/Tj    60 C | GaN HEMT:  10.0 W/Tj    68 C
# 100 kHz | Si SJ MOSFET:  21.0 W/Tj    75 C | Si IGBT:  49.5 W/Tj    99 C | SiC MOSFET:  12.8 W/Tj    65 C | GaN HEMT:  12.0 W/Tj    72 C
# 300 kHz | Si SJ MOSFET:  45.0 W/Tj   104 C | Si IGBT: 129.4 W/Tj   179 C | SiC MOSFET:  24.8 W/Tj    80 C | GaN HEMT:  16.9 W/Tj    81 C

At 20 kHz all four technologies are thermally comfortable and the cheapest adequate device wins. At 100 kHz the IGBT's switching energy dominates (≈50 W). At 300 kHz only SiC and GaN remain efficient; the SJ MOSFET is borderline and the IGBT impossible. The recommendation therefore depends on the stated frequency, efficiency target, cost and supply — not on a universal ranking.

Common mistakes

Applications

Key takeaways

Going further