Gate Driver: Design Rules and Process Limits
A gate driver sits between a controller that runs at a few volts and a power switch that may sit hundreds of volts above or below it. The isolation between the two is the defining feature of the part, and the way it is laid out on the board decides whether the driver survives the switching transients it is expected to cross.
Why a Gate Driver Needs Isolation
A high side switch in a bridge has its source or emitter moving with the switching node, so its gate drive has to be referenced to a potential that swings at the switching frequency. A bootstrap circuit can supply that shift for a short duty cycle, and it fails when the switch has to stay on for a long time.
A galvanic isolation barrier solves both problems: the drive signal crosses it, and the isolated supply has its own reference so the duty cycle is unlimited. That is why an isolated driver is used in motor drives, solar inverters and any topology where the switch is not referenced to the control ground.
The barrier also carries a safety function. Where the power circuit is connected to the mains, the barrier is what keeps the user side separated from it, and its rating is then a matter of the applicable standard rather than of circuit behaviour.
Isolation Technologies Compared
An optocoupler uses a light emitting diode and a photodetector. It is well understood, it has been used for decades, and its weakness is the propagation delay and the ageing of the emitter, which shifts the current transfer ratio over the life of the part.
A capacitive isolator passes the signal through a pair of high voltage capacitors made in the package. Its delay is short and its timing is tight, but it needs a carrier and a modulator, and it can produce emissions that a sensitive circuit nearby may notice.
A transformer based isolator, including the magnetic and the giant magnetoresistive types, gives the best common mode immunity because the coupling is magnetic rather than capacitive. The difference matters in a fast switching converter, where the isolation barrier sees a very fast voltage step.
common mode transient immunity
common mode transient immunity is the rate of change of the barrier voltage that the part can withstand without producing a wrong output. It is quoted in kilovolts per microsecond, and the value is a property of the isolation technology and of the internal design.
In a bridge driven at eight hundred volts with a fifty nanosecond edge, the switching node moves at more than ten kilovolts per microsecond, and a part rated at twenty five kilovolts per microsecond has only a modest margin. Derating the figure for temperature and for the production spread is normal practice.
The immunity is also direction dependent. A part may be specified at a high value for a transient that raises the secondary side above the primary and a lower one for the reverse. Checking both directions against the actual waveform is part of choosing the part.
dead time and Shoot Through
dead time is the interval during which both switches in a bridge are off, inserted so that the two never conduct at once. Without it, a momentary short across the supply destroys the devices, and the current rises at a rate set by the supply voltage and the loop inductance.
The driver usually has a programmable dead time, and the value has to be long enough for the slowest switch in the worst case and short enough to limit the distortion it introduces. A silicon carbide device turns off in tens of nanoseconds while an equivalent silicon part takes ten times as long, and the dead time has to accommodate whichever is fitted.
Adding dead time creates a non linearity at low output currents, because during the dead interval the output voltage is set by the load current rather than by the control. Compensating it in software is possible, and the minimum practical dead time is the better answer where the hardware allows.
desaturation protection and Soft Turn Off
desaturation protection watches the collector voltage of an insulated gate bipolar transistor while it is on. A voltage above a threshold means the device is not fully conducting, which happens during a short circuit, and the driver turns it off before it is destroyed.
The blanking time is the compromise in that scheme. The sense has to be ignored while the device turns on and the voltage falls, but it has to act quickly once a real fault occurs. A blanking time of a few microseconds suits most devices, and it is set from the datasheet switching time.
The turn off during a fault has to be slow. The current in a short circuit is several times the rated value, and turning it off at the normal speed produces a voltage spike large enough to destroy the device. A soft turn off path with a higher gate resistance is standard, and it must be checked at the lowest operating temperature where the fault current is highest.
Isolated Supply and Bias Voltage
The isolated side needs its own supply, usually a small transformer with a rectifier and a regulator. The supply has to be isolated as well as the signal, and its isolation rating has to match the barrier in the driver.
The gate voltage has to be right for the device. A silicon carbide device needs a negative off voltage to prevent a spurious turn on during a fast transition, while a silicon device may work with zero volts. The negative bias has to be generated in the isolated supply, which adds a winding or a charge pump.
The supply also has to source and sink the gate charge on every cycle. A gate charge of a few hundred nanocoulombs at twenty kilohertz is a continuous current of milliamps, and the decoupling at the driver has to supply the pulse without letting the rail sag.
Layout, Creepage and the Barrier
Keep the primary and secondary copper strictly on their own sides of the package, with no track crossing under the part, and respect the creepage and clearance distances for the voltage and the pollution degree. A slot under the barrier increases the effective creepage where the distance is large.
The gate loop is the other critical area. The gate trace and its return should form a small loop, because the inductance of that loop rings with the gate capacitance and slows the switching while producing emissions. A few nanohenries are enough to change the waveform noticeably.
The isolation barrier must not be bridged by a mounting hole, a thermal pad or a test point. A common failure in prototype boards is a ground plane that extends under the isolated side, which destroys the isolation while looking entirely reasonable. The general discipline for keeping two domains apart is described in our guide to mixed signal board design.
Verification and Faults
Verify the driver by measuring the gate waveform at the device terminals rather than at the driver output. The difference between the two shows the effect of the loop inductance, and a waveform that rings severely at the driver pin is usually fine at the gate.
Test the common mode immunity by applying the expected transient and watching for a wrong output. A bench test with a fast pulse generator across the barrier is worth more than the datasheet figure, because the real waveform includes ringing that the specification does not.
A driver that fails in the field usually does so after a switching event rather than during steady operation. Check the creepage, the isolation barrier under the part and the soft turn off behaviour. The release checks that catch these layout errors are collected in our PCB design release checklist, and the assembly points we inspect on a finished board are listed in judging PCB quality.
FAQ
Why does my isolated gate driver give a false output during switching? The common mode transient at the barrier exceeds the immunity of the part. Choose a driver with a higher rating and check the isolation layout.
How much dead time do I need? Long enough for the slowest device at the worst temperature, and as short as the resulting distortion allows.
Is a bootstrap supply enough? For a short duty cycle it can work. If the high side switch has to stay on indefinitely, an isolated supply is required.
