Fast GaN edges turn small PCB parasitics into serious gate-drive problems. Learn to separate real gate excursions from probing artifacts, select a device-compatible GaN gate driver, and validate Kelvin-source routing and gate-resistor tuning with double pulse testing.
A silicon half-bridge can appear well behaved while hiding layout weaknesses that become destructive after a move to GaN. Faster voltage and current transitions expose shared source inductance, gate-loop resonance, and capacitive coupling that were previously tolerable, and an oscilloscope can either reveal those problems or create a convincing imitation of them. Successful migration therefore requires treating the transistor, driver, PCB, and measurement setup as one switching system. [1][3][4]
650 V GaN is relevant to compact converters operating in the 400 V bus class, but the voltage-class label does not establish an acceptable operating bus or transient margin. Actual bus tolerance, switching overshoot, and device limits still have to be reconciled for the selected component. The supplied power-development guide identifies GaN gate sensitivity and driver immunity to high dv/dt as particular concerns in this application class. [6]
More importantly, “GaN” does not define a universal gate interface. The driver-selection research explicitly warns that gate-voltage tolerances and drive requirements vary between GaN technologies and manufacturers. A driver that works with one device may be unsuitable for another even when both components are marketed for GaN power conversion. [3]
The broad GaN overview illustrates the danger of generalizing: it describes a 0 to 6 V drive arrangement and a transient-tolerant −20 V to +10 V gate range in its discussion. Those numbers must not become requirements for an unrelated transistor. They are not universal permissions to drive every enhancement-mode GaN gate to those voltages. [1][3]
Before editing the schematic, create a device-specific gate-drive envelope containing:
Keep recommended operating values separate from absolute maximum ratings. A transient rating is not a target waveform, and threshold voltage is not a guaranteed boundary between harmless behavior and destructive cross-conduction.
A useful troubleshooting model separates three mechanisms: shared-source voltage, gate-loop resonance, and displacement current from the switching node. They interact, but each suggests a different corrective action. The GaN overview identifies stray inductance and gate-drain coupling as mechanisms behind ringing, false switching, and possible failure. [1]
Common-source inductance is the portion of the source connection shared by the power-current path and gate-drive return. Its induced voltage follows:
v_L = L_common × di/dt
The driver controls its output relative to its own return terminal. The transistor responds to gate voltage relative to its local source. When those reference points separate electrically during a fast current transition, the commanded driver voltage and actual device gate-source voltage are no longer identical. [1][3]
The sign of the disturbance depends on current direction and the switching event. Shared-source inductance can oppose a commanded transition, produce source bounce, or participate in feedback that sustains ringing. It should not be modeled as a fixed positive gate spike in every operating condition.
This distinction is especially important when investigating a supposedly clean gate waveform measured at the driver. A quiet driver output does not prove that the transistor sees a quiet gate-source voltage.
The corrective action is to remove shared impedance, not simply to increase driver strength. Use the device's Kelvin-source connection where available. Otherwise, follow the manufacturer's recommended source pickup and connect the gate return at the device rather than downstream along a power-current trace. A board-level Kelvin route cannot eliminate inductance already shared inside a package. [1][3]
The gate loop includes the driver output, gate resistor, gate interconnect, device input structure, source return, and the relevant local supply-decoupling path. Parasitic inductance interacting with effective gate capacitance can produce overshoot and undershoot even when the driver supply itself is correctly regulated. [1][3]
A simplified resonance estimate is:
f_r ≈ 1 / (2π√(L_gate × C_effective))
This is a diagnostic model, not a complete transistor model: device capacitances vary with operating conditions, and driver output impedance is not an ideal resistor. Nevertheless, it explains why increasing physical loop area can change ringing frequency, and why adding resistance can reduce ringing while also slowing switching.
A large source/sink current specification is therefore not sufficient evidence of a good driver match. The research emphasizes both current capability and parasitic impedance; strong current drive through an inductive loop can still produce unacceptable gate excursions. [1][3]
The intrinsic gate-drain capacitance provides a coupling path from drain voltage movement into the gate. External PCB capacitance can add another path when switch-node copper lies near gate traces, driver nodes, or other sensitive conductors. The instantaneous displacement-current relationship is:
i_coupled = C_coupling × dv/dt
That current must leave through the off-state gate network. Its voltage effect depends on the network's frequency-dependent impedance, including driver sink impedance, gate resistance, and inductive return paths. Low Miller charge does not make a device immune to high-dv/dt disturbance. [1]
This is why a compact gate loop and a strong, local off-state return are complementary requirements. A low-resistance pull-down located far away may still present substantial impedance during the fastest part of a transition.
False turn-on and gate overvoltage can occur together, but they are different problems. False turn-on means a nominally off device begins conducting unintentionally. Gate overvoltage means the gate experiences a voltage outside its permitted envelope, whether or not that excursion causes immediate conduction. [1][3]
A third possibility is measurement error. The probing research documents false ringing from inductive ground leads, common-mode limitations in differential probes, and waveform changes caused by probe capacitance. Although that article discusses SiC, these measurement mechanisms also apply to fast GaN switching nodes. [4]
| Observation | Plausible mechanism | Discriminating check | Likely design response |
|---|---|---|---|
| Off-device gate disturbance coincides with opposite-device switching | Intrinsic or PCB capacitive coupling | Correlate local VGS with switch-node dv/dt and current | Reduce coupling and off-state impedance; tune the aggressor's edge |
| Gate excursion follows rapid current change | Shared-source inductance | Compare driver-referenced and device-referenced measurements | Improve Kelvin pickup and separate gate return from power current |
| Gate rings after its own drive transition | Gate-loop resonance | Change loop geometry or damping while holding operating conditions constant | Shorten the loop and tune device-approved gate resistance |
| High-side gate spike changes strongly with probe arrangement | Common-mode conversion or pickup | Repeat with a suitable isolated measurement system | Correct the measurement before changing gate bias |
| Behavior changes when the gate probe is removed | Probe loading | Compare switching behavior using lower-loading instrumentation | Validate the final network without relying on probe capacitance |
| Current spike appears without credible evidence of gate conduction | Capacitive commutation, sensing artifact, or another mechanism | Correlate current with both gate waveforms and drain voltage | Avoid diagnosing shoot-through from current alone |
These checks form a hypothesis-driven workflow rather than a one-to-one fault lookup. Several mechanisms can coexist, and the measurement setup can alter the circuit being diagnosed. [1][4]
Measure VGS between the gate connection and the relevant local source reference. Do not substitute gate-to-system-ground voltage for high-side VGS: the high-side driver and gate ride on a moving source potential. The driver-selection guide explicitly identifies this source-referenced requirement. [3]
For high-side measurements, evaluate all of the following:
A high-voltage differential probe can have adequate voltage range yet insufficient common-mode rejection to resolve a relatively small gate signal on a rapidly moving switching node. The probing research identifies isolated probes as a better alternative for demanding high-side gate measurements. [4]
The familiar relationship BW ≈ 0.35 / rise_time is useful for an initial bandwidth estimate. The probing article gives 35 MHz for a 10 ns transition on a 1,200 V SiC device. That is an illustrative measurement-system relationship, not a recommended bandwidth for all GaN gate measurements. [4]
Resolving narrow overshoot or ringing can require more bandwidth than recognizing the overall transition. Conversely, additional displayed bandwidth can reveal probe pickup rather than a real device waveform. Evaluate the entire measurement chain: tip, connection, probe, oscilloscope, and common-mode performance.
Avoid long ground leads and improvised current-probe loops. The supplied probing reference reports that a wire loop introduced for current measurement can add approximately 20 nH per inch, and specifically warns that three-inch passive-probe ground leads create apparent ringing. [4]
Probe capacitance can slow gate transitions or damp ringing. The probing reference describes a particularly dangerous sequence: a loaded gate appears slow, the engineer reduces gate resistance, and removing the probe exposes ringing or unexpected turn-on. [4]
Use a controlled verification sequence:
For hazardous-voltage testing, use an appropriately rated enclosure, interlocks, probes, and accessories. Never lift oscilloscope protective earth or use an isolation transformer to float a conventional oscilloscope; the probing reference explicitly warns against that practice. [4]
Driver selection should begin with the transistor datasheet, not the controller's logic supply or an attractive peak-current headline. Relevant criteria include output-voltage compatibility, source and sink capability, propagation delay, dead time, high-side operation limits, noise immunity, and PCB parasitics. [3]
Consider the regulated driver supply together with its tolerance, startup behavior, local decoupling, bootstrap behavior where applicable, and transient gate ringing. The nominal supply can satisfy the recommended gate voltage while the actual gate waveform violates a limit. Tight supply regulation does not replace a low-inductance gate loop. [1][3]
For the high-side channel, also check operation during switch-node excursions and rapid common-mode transitions. A compatible output voltage alone does not establish immunity to false commands or disturbed driver operation. [3][6]
The approximate charge relationship I_average ≈ ΔQ / Δt helps estimate the current needed for a desired transition. It is only a starting point: internal driver impedance, external resistance, and parasitic inductance determine the actual current waveform. The driver guide explicitly recommends checking both sourcing and sinking capability. [3]
Independent turn-on and turn-off paths are useful because the objectives differ. Turn-on impedance can control the aggressor's dv/dt and di/dt, while the off-state path must hold the complementary device near its intended off voltage. The GaN overview recommends independent gate resistors for tuning these behaviors. [1]
Do not promote a generic resistor range into a production value. Begin with the selected device and driver reference design, then validate on the intended package, stackup, and layout.
Include undervoltage, controller reset, missing commands, and power sequencing in the review. Confirm what the output does when the driver is unpowered or transitioning through its supply thresholds, and whether the external network maintains the intended off state. Device compatibility and required protection functions are part of the driver-selection process, not optional additions after efficiency testing. [3]
Place local DC-link decoupling to minimize the high-current switching loop through the half-bridge. The supplied layout guidance recommends locating input and decoupling capacitors just millimeters from the switching devices and notes the potential benefit of vertically arranged multilayer power loops. [2]
Reducing this loop's inductance addresses drain-voltage overshoot and ringing at their source. It also reduces the severity of the electromagnetic environment around the gate circuitry.
Place each driver close to its transistor. Route the gate and its source return as a tightly coupled pair, and place local driver decoupling so that fast drive current does not take a long detour through the supply network. The GaN research recommends short, low-inductance interconnects and close driver placement. [1][2]
Do not optimize trace width in isolation. Loop area and return-path continuity matter. A wide gate trace separated from its return by a large area can still produce a poor high-frequency loop.
Return the driver to the designated Kelvin-source pin when one exists. Keep this connection separate from the path carrying switching current. For packages without a dedicated Kelvin terminal, use the manufacturer's preferred local source pickup rather than assuming any point on the source net is equivalent. [1][3]
Keep switch-node copper away from gate traces and sensitive driver nodes. The objective is not merely visual separation: it is reducing unwanted capacitive coupling. Use appropriate local reference structures, recognizing that the high-side driver reference is the moving source node, not necessarily the system ground plane. [1][3]
Finally, design probe access into the layout. A board that can only be measured through long flying leads makes reliable validation unnecessarily difficult. Test-point geometry, bandwidth, and voltage rating are part of the power-stage design. [4]
Begin with the device vendor's recommended network and a reduced-energy test condition. Adjust the turn-on resistance while observing switch-node slew, off-device VGS, current, and switching loss. Then assess the turn-off path and ringing. Independent gate resistors help separate these objectives, although changing one edge can still affect the opposite device. [1]
Increasing resistance can reduce slew and add damping, but excessive turn-on resistance increases switching time and loss. Conversely, reducing turn-off resistance cannot compensate indefinitely for a long inductive return path. The physical loop must be corrected before expecting resistor tuning to provide robust margin. [1][3]
The GaN overview lists −3 V turn-off bias among possible mitigation techniques. The driver-selection research, however, makes clear that acceptable drive parameters vary by device. Treat negative bias as an option requiring explicit device approval, not as a standard GaN upgrade. [1][3]
Before applying it, verify:
Do not confuse negative off bias for a normally-off power switch with the bias required to turn off a normally-on device. The supplied RF GaN discussion emphasizes that depletion-mode devices have distinct bias-sequencing requirements; material name alone does not make their drive circuits transferable to switching-power devices. [5]
The broad GaN overview mentions ferrite beads and gate-source RC networks as possible ringing mitigations. These are candidates for evaluation, not universally safe fixes. [1]
A ferrite changes frequency-dependent gate impedance. A gate-source capacitor or RC network changes drive loading and timing. A clamp has its own voltage tolerance, capacitance, dynamic behavior, and connection inductance. Evaluate whether the protection network actually limits voltage at the device, rather than at a remote point on the PCB.
Keep gate protection distinct from bridge fault protection. A gate clamp does not prevent every form of shoot-through, and dead time does not prevent dv/dt-induced turn-on of a device already commanded off. Driver timing and noise-induced switching require separate checks. [1][3]
Double pulse testing provides switching events at a controlled target current while helping manage device temperature. It is identified in the supplied research as a tool for switching-loss evaluation and SiC/GaN gate-drive tuning. [4][6]
Document which device is pulsed, how the inductor is connected, and where current flows between pulses. In the usual arrangement, the first pulse establishes inductor current and the second creates a switching event at that current. The complementary device and its off-state gate network remain part of the experiment; their behavior must not be treated as an ideal background condition.
For an approximately constant inductor voltage, ΔI ≈ V_L × Δt / L provides a first estimate of the pulse duration. Confirm the actual current waveform rather than relying solely on the calculated value.
Record, using suitably rated instrumentation:
Account for probe delay when comparing timing or calculating switching energy. Because measurement capacitance and current-sensor inductance can alter the switching event, document those contributions alongside the result. [4]
Start with reduced bus voltage and pulse current, then expand the test matrix toward intended operating conditions. Include both switching directions where the converter topology requires them, relevant temperature conditions, gate-supply variation, dead-time variation, and the final component network. Driver timing, voltage compatibility, and parasitics must be evaluated together. [3][4]
Predetermine acceptance criteria: permitted gate envelope, acceptable drain-voltage stress, absence of credible unintended conduction, and acceptable switching loss. Do not declare false turn-on solely because a displayed gate spike crosses a typical threshold value. Correlate a credible local VGS measurement with current and voltage behavior, and exclude capacitive commutation and measurement artifacts.
Likewise, a low-energy double-pulse pass does not establish startup behavior, fault response, or repetitive thermal performance. Follow it with controlled converter operation and checks of the supply and protection states identified during driver selection. [3][6]
A fast switch-node transition pushes displacement current (i = C × dv/dt) through the gate-drain capacitance and any PCB coupling into the gate of the device that is supposed to stay off. Whether that becomes unintended conduction depends on the impedance of the off-state gate network. Shared-source inductance and gate-loop resonance can add to the disturbance, so each mechanism needs its own check. [1]
Return the driver to the device's Kelvin-source pin when one exists, and otherwise follow the manufacturer's recommended source pickup. This keeps the gate return out of the power-current path. A board-level Kelvin route cannot remove inductance that is already shared inside the package. [1][3]
Measure VGS at the device against its local source, use an isolated probe for demanding high-side measurements, and avoid long ground leads, which the probing reference shows can create apparent ringing. If the waveform changes when you reroute or remove the probe, correct the measurement before changing the gate network. [4]
Only when the selected device's vendor approves it. Negative bias such as −3 V appears among possible mitigations, but acceptable drive voltages vary between GaN technologies and manufacturers, and worst-case undershoot adds to the commanded bias. [1][3]
Double pulse testing creates switching events at a controlled current while limiting device heating, which makes it useful for switching-loss evaluation and gate-drive tuning. Capture VGS on both the switched and the off device, along with drain voltage and current. A pass does not cover startup, fault response, or repetitive thermal performance. [4][6]
AI-driven electronics workflows such as IntelCAD.ai can help turn these requirements into a reviewable schematic and PCB starting point: separate gate resistors, local driver decoupling, explicit Kelvin returns, and planned measurement access. They can also help organize device constraints and maintain a consistent troubleshooting checklist as the design changes.
The useful boundary is clear: automation can support requirements capture and design review, but it cannot certify a nanosecond gate waveform from connectivity alone. Final confidence still comes from the selected vendor's limits, careful physical implementation, and measurements that have themselves been validated.
Design your next board by describing it.
IntelCAD.ai turns plain-language requirements into schematics and PCBs.
Start Your Design