Electronics Guide

Bandgap and Critical Electric Field

Silicon has a bandgap near 1.1 electron volts. The 4H polytype of silicon carbide is close to 3.3 electron volts, and gallium nitride is slightly wider still, near 3.4 electron volts. A wider bandgap makes avalanche breakdown harder to trigger, so the material tolerates a much stronger electric field before it conducts uncontrollably: roughly 0.3 megavolts per centimeter for silicon against something near ten times that for both wide-bandgap materials.

That factor of ten is the whole story in compressed form. To block a given voltage, the drift region can be roughly ten times thinner and doped roughly a hundred times more heavily. Because on-resistance scales with thickness divided by doping, the theoretical resistance per unit area for a given blocking voltage falls by two to three orders of magnitude. Real devices never capture the full theoretical gain—channel resistance, substrate resistance, and manufacturing constraints intervene—but the direction and rough magnitude hold.

Thermal Conductivity and Operating Temperature

Silicon carbide conducts heat roughly three times better than silicon, which is a significant practical advantage when a smaller die must dissipate comparable power. Gallium nitride is the weaker performer here, and commercial GaN power transistors are lateral devices grown on foreign substrates—usually silicon, sometimes silicon carbide or sapphire—so heat must travel sideways to contacts and then through an interface layer. That is one reason GaN has concentrated at lower voltages and smaller die sizes, where the heat to be removed is modest.

Wide-bandgap materials also retain low leakage at high temperature, so the intrinsic material limit sits far above 200 degrees Celsius. In practice the limit is rarely the semiconductor. Solder, die-attach material, encapsulant, and gate dielectric fail first, and most commercial devices are still rated for maximum junction temperatures of 150 to 175 degrees Celsius. Raising that ceiling is a packaging problem more than a materials problem.

Unipolar and Bipolar Conduction

A second axis, independent of material, is whether the device conducts with one carrier type or two. Unipolar devices—MOSFETs, HEMTs, Schottky diodes—carry current with majority carriers alone. They behave resistively, switch quickly, and store no minority-carrier charge, so they have no reverse-recovery penalty and parallel readily because on-resistance rises with temperature.

Bipolar devices—IGBTs, thyristors, PIN diodes—inject minority carriers into the drift region. This conductivity modulation collapses the on-state voltage drop, which is what makes thick, high-voltage drift regions usable at all. The cost is stored charge that must be removed at turn-off, producing tail current in IGBTs and reverse-recovery current in diodes, both of which limit switching frequency. Silicon's low critical field forces bipolar conduction above roughly 1,000 volts; wide-bandgap materials push the unipolar option far higher, which is precisely why SiC MOSFETs displaced silicon IGBTs in the 1,200-volt class.

Below 200 Volts

Silicon MOSFETs remain dominant and inexpensive here, serving point-of-load converters, low-voltage motor drives, and battery-powered equipment. Gallium nitride transistors in the 100-volt class compete where switching frequency or size matters more than device cost, notably in 48-volt data-center architectures and class-D audio.

The 600 to 900 Volt Class

This is the most contested territory in power electronics, covering single-phase mains rectification, solar string inverters, on-board vehicle chargers, and server power supplies. Silicon superjunction MOSFETs, 650-volt GaN HEMTs, and 650-volt SiC MOSFETs all compete, and the choice usually turns on switching frequency and system cost rather than on device efficiency alone. GaN generally wins above a few hundred kilohertz because it has the lowest gate charge and output charge and no body-diode reverse recovery; silicon superjunction devices remain the cost benchmark at lower frequencies.

The 1,200 to 1,700 Volt Class

Industrial motor drives, commercial solar inverters, and electric-vehicle traction inverters live here. Silicon IGBTs held this range for decades and still serve cost-sensitive designs. Silicon carbide MOSFETs have taken a large share of automotive traction, especially in vehicles with 800-volt battery architectures, where the reduction in switching loss allows higher switching frequencies, smaller filters, and simpler cooling. The 1,700-volt class serves 690-volt industrial systems, traction auxiliaries, and grid-tied converters. Gallium nitride has begun to reach this class as well, with 1,200-volt parts in production aimed principally at high-voltage direct-current distribution inside data centers.

3.3 Kilovolts and Above

Rail traction, medium-voltage drives, wind-turbine converters, and grid equipment use IGBT modules and press-pack devices at 3.3, 4.5, and 6.5 kilovolts. Integrated gate-commutated thyristors compete in the same range with lower conduction loss and higher fault-current capability, at the price of bulky gate-drive units and turn-on snubbers. Silicon carbide devices at 3.3 kilovolts have reached commercial availability, and 6.5- and 10-kilovolt SiC devices have been demonstrated in research and pilot production, but they remain expensive and are not yet mainstream.

Utility Scale

At the top of the range, phase-control thyristors are fabricated one device per wafer, on wafers up to 150 millimeters in diameter, with blocking voltages of roughly 8 kilovolts and average currents of several kiloamperes. No other semiconductor approaches their surge-current capability or their conduction efficiency. Line-commutated HVDC converters stack roughly a hundred such devices in series per valve to reach transmission voltages of 800 kilovolts and beyond. Voltage-source HVDC converters instead use large numbers of series-connected IGBT submodules, trading some efficiency for the ability to control real and reactive power independently and to energize passive networks.

Reading the Voltage Rating

A blocking-voltage rating is an absolute maximum, not an operating point. Switching transients across parasitic inductance add overshoot to the DC-link voltage, so designers typically operate at 50 to 70 percent of the rating. At high altitude and in equipment expected to run for decades, terrestrial cosmic radiation imposes a further constraint: high-energy neutrons can trigger destructive single-event burnout in a blocking device, at a failure rate that climbs steeply with applied voltage. Manufacturers publish failure-in-time curves against DC voltage, and high-reliability designs derate against them explicitly.

Conduction and Switching Loss

Total device loss splits into conduction loss and switching loss, and the two respond differently to design choices. Conduction loss depends on current and the device's on-state characteristic: a resistive drop for unipolar devices, which rises with temperature, or an offset voltage plus a small resistance for bipolar devices, which is comparatively flat with temperature. Switching loss is energy per transition multiplied by frequency, so it dominates as frequency rises.

This split explains most selection decisions. Below a few kilohertz, conduction loss dominates and bipolar devices win. Above roughly 20 kilohertz, switching loss dominates and unipolar devices win. The crossover frequency is exactly what wide-bandgap materials moved upward, and datasheet switching-energy figures—measured under specific gate resistance, temperature, and bus voltage—are the numbers that decide it.

Safe Operating Area and Short-Circuit Withstand

Beyond steady-state ratings, a device must survive defined excursions. Reverse-bias safe operating area bounds the simultaneous voltage and current permitted during turn-off. Short-circuit withstand time specifies how long the device can carry a fault before the gate drive must intervene. A silicon IGBT typically survives about 10 microseconds; a silicon carbide MOSFET, with its much smaller die area and higher current density, typically survives only 2 to 5 microseconds. That difference forces faster desaturation detection and soft-shutdown circuitry in SiC gate drives, and it is a common source of trouble when a design is ported from IGBTs to SiC.

The Thermal Path

Junction temperature, not current, is the real limit. Datasheet current ratings assume a specified case temperature that a practical cooling system may not achieve, so the useful calculation runs from computed loss through the thermal impedance network—junction to case, case to heatsink, heatsink to ambient—to a predicted junction temperature. Transient thermal impedance curves matter as much as steady-state values for pulsed and cyclic loads, where the die heats and cools within each fundamental cycle.

Module Construction

Above a few tens of amperes, devices are supplied as modules rather than discrete packages. A typical module solders one or more dies to a ceramic substrate of direct-bonded copper, using alumina, aluminum nitride, or silicon nitride as the insulator; the substrate is then attached to a baseplate, and aluminum wire bonds carry current from the die top surface. Silicon nitride substrates and silver-sintered die attach are increasingly used in demanding applications because they survive far more thermal cycles than the conventional alumina and solder combination. Press-pack construction, in which dies are clamped rather than soldered, avoids bond wires entirely and fails short-circuit rather than open, which is what allows series stacks to keep operating after a device failure.

Parasitic Inductance

Loop inductance in the commutation path converts fast current transitions into voltage overshoot, and its importance grows in direct proportion to switching speed. A layout adequate for a silicon IGBT switching at 1,000 amperes per microsecond can destroy a silicon carbide device switching ten times faster. Low-inductance module designs use laminated busbars and interleaved terminals, and decoupling capacitors are placed as close to the die as construction allows. Common-source inductance shared between the power loop and the gate loop is particularly damaging, since it feeds a voltage opposing the gate drive back into the gate circuit and slows the very transition the designer is trying to accelerate; devices intended for fast switching provide a separate Kelvin-source connection to eliminate it.

Paralleling and Series Connection

Paralleling extends current capability but requires that devices share. Unipolar devices with positive temperature coefficients of on-resistance share static current naturally, since a device carrying more current heats, becomes more resistive, and sheds current. Dynamic sharing during switching is the harder problem and depends on symmetric layout and matched gate-drive impedance. Series connection extends voltage capability but requires deliberate voltage balancing, using resistor networks for static distribution and snubbers or active gate control for dynamic distribution—the technique on which HVDC valves depend.