GaN inverter design.
For efficient motor drives.
We develop GaN HEMT inverters for PMSM and BLDC motors. Our design work addresses power loss, cooling and current control to improve efficiency and output capability.
MAX 80 V DC250 A rms · phase
POSITION FEEDBACK OPTIONS UNDER REVIEW
We review the power stage, cooling structure and control requirements against device characteristics and operating conditions.
GaN HEMT INVERTERThe following sections explain design principles and review criteria. They do not report product test results or confirmed specifications.
Fast transitions can reduce switching loss, but high current slew rates can also produce voltage overshoot across parasitic inductance. Device selection and PCB layout need to be considered together.
Gate voltage and drive impedance must suit the selected device. Power loops, gate loops and return paths influence ringing, unintended switching and electromagnetic emissions.
DESIGN REVIEW POINTS
Conduction and switching losses are major sources of heat in the power stage. Loss estimates depend on current, switching frequency and device temperature; thermal design must also account for the path from the device to the cooling system.
Continuous operation and short peak loads produce different thermal responses. Estimating peak junction temperature requires load duration, repetition and transient thermal response as well as average loss and steady-state thermal resistance.
DESIGN REVIEW POINTS
Simplified estimates for understanding a single device.
P_cond approximately equals I_rms squared times R_DS on
At constant on-resistance, conduction loss is proportional to RMS current squared. Iᵣₘₛ is the on-state channel current of one device evaluated over the full period, with off intervals set to zero. It differs from motor phase RMS or RMS over only the on interval; do not apply duty again. In practice, RDS(on) varies with temperature and other operating conditions.
P_sw approximately equals f_sw times the sum of E_on and E_off
For fixed Eon and Eoff, switching loss increases with switching frequency. The equation assumes one turn-on and turn-off pair per period. Both energies depend on operating voltage, current and gate-drive conditions.
Total inverter loss requires a separate calculation. Check the scope of measured loss terms to avoid omissions or double counting.
Motor control and overcurrent protection require current measurements at appropriate sampling instants. Switching noise, signal settling, sensing topology, PWM timing and ADC latency determine the usable measurement window.
Dead time is used to avoid simultaneous conduction of the upper and lower switches in the same leg, but also affects reverse-conduction loss. Gate-drive delays and switching timing must be reviewed together. Fault detection and shutdown response require a separate timing assessment.
DESIGN REVIEW POINTS
FROM DC POWER TO MOTOR DRIVE
The inverter switches a DC supply to produce motor terminal voltages. Current measurements and motor feedback inform control and protection design.
Conceptual energy flow. Not a product configuration or specification.
WHY GaN / BEYOND THE DEVICE
Gate charge, reverse conduction and switching frequency affect power-stage losses and operation. Evaluate each under the intended circuit and operating conditions.
Low gate charge can support fast transitions and lower gate-drive loss.
Actual transition speed depends on the device, driver and circuit parasitics.
Gate-drive principles · TI (opens in a new tab)E-mode GaN has no p–n body-diode reverse-recovery charge.
Reverse conduction and output-capacitance charging currents still occur. Review dead-time voltage drop and duration separately from capacitance-related switching loss.
GaN reverse conduction · TI (opens in a new tab)Fast switching provides greater flexibility in selecting the switching frequency.
Review transition and drive losses, current ripple and emissions together. Output-capacitance charging and its associated losses also depend on the circuit and switching conditions.
GaN loss mechanisms · TI (opens in a new tab)ENGINEERING / LOSS & HEAT
Compare semiconductor conduction and switching losses with magnetic core and winding losses, then examine how heat flows to the cooling system.
LOSS ATLAS / DEVICE TO SYSTEM
Compare models for conduction, switching and magnetic-component losses, including the boundaries and conditions needed to calculate total inverter loss.
The boundary here is the inverter. Magnetic losses apply only to inductors or transformers actually inside that boundary. Motor iron and copper losses are separate. Align time intervals and measurement boundaries to avoid double counting.
j: semiconductor device · m: magnetic component · cond: on-state conduction · sw: transitions · rev: separate reverse conduction · core/cu: core/winding · aux: remaining drive/control/supply losses not counted above · other: capacitors, interconnects and remaining losses
CONDUCTION / SENSITIVITY
GaN conduction loss depends on the current waveform, conduction intervals, junction temperature, gate-drive conditions and preceding voltage stress. These conditions must also be considered when applying an I²R model.
In the resistive on-state, treat RDS(on) as an effective resistance dependent on Tj, VGS, device current i and stress/recovery history h. The mask g(t) is 1 during on-state channel-conduction intervals and 0 otherwise. Exclude transition intervals and off-state reverse conduction.
Dynamic on-resistance in GaN can differ from the DC value. Characterization must match off-state voltage, stress duration, switching conditions and observation time. The graph intentionally omits these dependencies to isolate the I²R relationship.
Tobs [s]: averaging interval · i [A]: device current · RDS(on) [Ω] · Tj: junction temperature · VGS [V] · g: dimensionless. At fixed resistance, I²rms,on = (1/Tobs)∫g(t)i²(t)dt: full-interval RMS with off intervals set to zero. Conduction duty is already included; do not multiply by duty again or directly substitute motor phase-current RMS.
SWITCHING / SENSITIVITY
Switching frequency alone does not determine switching loss. Evaluate the voltage, current, temperature, drive conditions and parasitics at each device transition.
Sum across all devices j and transition events k within the observation interval. Eon,k and Eoff,k are loss energies at their respective operating points. Relate the terminal integral ∫vDS iD dt to heat only after accounting for stored/recovered energy and measurement boundaries.
E = E(VDS, ID, Tj, VGS, Rg, parasitics, commutation mode). Align voltage/current probe timing and integration windows, and use the actual hard- or soft-switching conditions.
Evaluate reverse conduction during dead time from its voltage drop, current and duration. An e-mode GaN HEMT has no intrinsic p–n body-diode reverse recovery; this does not imply zero reverse-conduction or output-capacitance-related loss. Assess cascodes and external diodes separately. Do not add a term already included in Eon/Eoff.
Tobs [s] · E [J] · P [W] · fsw [Hz]. For one device, one turn-on and one turn-off per cycle at fixed conditions give Psw ≈ fsw(Eon + Eoff). Ωrev,j contains gate-off reverse-conduction intervals of device j, excluding on-state g(t) intervals and transition-energy windows.
MAGNETIC / SENSITIVITY
The total loss of a magnetic component is the sum of core loss and winding loss. Core loss depends on the flux-density waveform B(t); winding loss depends on DC and harmonic currents and resistance at each frequency.
The sinusoidal Steinmetz coefficients k, α and β depend on material, temperature and the fitted frequency/flux range. Here Bpk is the peak of a sinusoid without DC bias, so ΔB = 2Bpk. For PWM flux, use the actual B(t) slopes and excursions; substituting switching frequency alone is insufficient.
This iGSE describes a single closed flux loop per period. ΔB is that loop’s peak-to-peak flux density and TB is its period. ki is converted using the chosen Steinmetz convention; it is not k. Separate major and minor loops, sum all loop loss energies, and divide by the complete observation period.
DC bias, temperature, relaxation during zero-voltage intervals and local flux concentration are not generally resolved by the basic iGSE. Use measurements within the material/waveform range and extended models where needed. Do not add separately decomposed hysteresis/eddy-current losses on top of the total core loss fitted by iGSE.
For linear periodic steady state, use each harmonic RMS current and AC resistance. Rac is the total resistance for a specified winding/return-field condition, including DC resistance, skin and proximity effects. Significant multiwinding phase relationships or gap fringing require the corresponding field or resistance-matrix model. Do not add Rdc·Irms,total² again.
pcore [W/m³] · Vcore [m³] · Pcore [W] · B, ΔB [T] · TB [s] · α, β: dimensionless · k, ki [W·m−3·sα·T−β]. Numerical coefficients must match the f [Hz], B [T] and pcore [W/m³] unit convention. In,rms [A], Rac [Ω], f and f0 [Hz]. f0 is the winding-current fundamental frequency; n is a positive integer. T in the winding equation is temperature on the resistance data’s scale. Skin depth δskin [m] uses a classical homogeneous-conductor approximation, with μ [H/m] and σ [S/m]. Vcore·pcore assumes approximately uniform loss density.
Each graph uses a different loss reference. The normalized curves cannot be added to calculate total loss or efficiency. Actual prediction requires device, magnetic-material, winding and operating-point data.
The package, PCB and cooling structure determine how heat leaves the device. To estimate junction temperature, account for power loss, thermal resistance and the temperature of the ambient air or coolant.
This steady-state approximation represents heat flow from one device, dissipating Ploss [W], through one equivalent path. ΔT [K] is the junction-to-ambient or junction-to-coolant difference; Rθ [K/W] must span the same two points. Do not multiply total inverter loss by one device’s Rθ. Mutual heating, parallel paths and transient response require separate analysis.
Conceptual relationship · not product measurements
Steady-state thermal resistance and temperature rise · Infineon (PDF, p. 6) (new tab)HEAT TRANSFER / FROM MATERIAL TO SYSTEM
Power dissipation generates heat. Junction temperature follows from the heat-flow path and its cooling boundary. Conduction through materials, temperature drops at contacts and convection into a fluid connect the heat source to the cooling system.
For heat flowing through a uniform flat layer, greater thickness increases thermal resistance. Greater conductivity and cross-sectional area reduce it.
Steady, one-dimensional conduction; constant k and A; no heat generation within the layer. L is thickness [m], k conductivity [W/(m·K)], and A area normal to heat flow [m²]. Q̇ is heat-transfer rate [W]; R is thermal resistance [K/W].
Surface roughness, flatness and clamping pressure affect thermal contact. Bulk conduction through a TIM and contact resistance at its two faces are separate terms.
hc is areal contact conductance [W/(m²·K)]; Act is the contact area to which it applies [m²]. Do not add the same contact resistance again to measured TIM resistance that already includes it.
The surface-to-fluid temperature difference drives heat transfer. Fluid properties, velocity, channel geometry and flow regime determine h.
A lumped approximation using h [W/(m²·K)] and representative surface and bulk-fluid temperatures. Wetted area Awet [m²] is distinct from TIM contact area.
Radiation depends on the difference of fourth powers of absolute temperature. When a surface loses heat by both convection and radiation, add their heat-transfer rates.
An isothermal diffuse-gray surface viewing only much larger isothermal surroundings, with view factor 1 and a nonparticipating medium. Arad is radiating area [m²]; ε is emissivity; Ts and Tsur must be in K. σ ≈ 5.6704 × 10⁻⁸ W/(m²·K⁴).
The net heat-transfer rate into the coolant depends on its mass flow rate, specific heat capacity and temperature rise. For the same absorbed heat rate and specific heat capacity, a higher mass flow rate gives a smaller inlet–outlet temperature difference.
Steady single-phase flow with constant cp; neglect shaft work and kinetic/potential energy changes. ṁ [kg/s], cp [J/(kg·K)], and Q̇coolant [W], the net heat-transfer rate entering the coolant.
FIELD EQUATION → THERMAL CIRCUIT
Conduction, convection and radiation become internal transport and boundary conditions in the heat equation. Reducing the field to temperature nodes, resistances and capacitances gives a thermal circuit model.
ρ is density [kg/m³], cp specific heat [J/(kg·K)], k isotropic thermal conductivity [W/(m·K)], and q‴ volumetric heat generation [W/m³]. With no phase change or thermomechanical coupling, the local energy-storage rate equals net conduction input plus internal generation, all per unit volume.
q″ = −k∇T [W/m²] is heat flux. Convection and radiation act as surface boundary conditions; contact interfaces have a heat-flux-dependent temperature jump. Specify the initial temperature and each boundary condition.
For approximately uniform node temperature and volumetric heat capacity ρcp, Cθ = ρcpV [J/K], where V is node volume [m³]. The part of power loss P(t) [W] not yet leaving through the boundary increases internal energy. Tb is the constant reference-boundary temperature; t is time [s].
With constant Rθ and Cθ, T(0) = Tb and constant Pstep applied from t = 0, the step response below follows. τ = RθCθ [s] sets the timescale. Cθ does not change the steady-state rise.
Resolve a large temperature variation with multiple nodes. Distinguish a network representing physical heat paths from a fitted Foster network; its internal nodes do not represent literal layer interfaces.
Models for coordinate transforms, current control and thermal response.
Closely coupled opposing current paths in the same loop partially cancel their magnetic fields, reducing parasitic loop inductance. Keep local DC-link connections to the switches short, while also considering switching-node parasitic capacitance.
Cross-section of the outgoing and return current paths
EPC power-loop layout reference (opens in a new tab)We review supply voltage, motor characteristics, cooling and control requirements to define the inverter’s application scope and development priorities.
Detailed specifications have not yet been released. The stated design targets are not test results or confirmed ratings.
ENGINEERING THE NEXT DRIVE
Supply and motor characteristics, cooling conditions and control requirements define the design inputs. We evaluate conduction and switching losses, thermal limits and current-sensing constraints to set efficiency, power-density and control targets.
Engineering concept image
Engineering concept image
These are development objectives and design considerations. Specifications, supported functions and evaluation conditions are agreed for each product; performance must be established by testing under those conditions.
CO-DEVELOPMENT / FROM IDEA TO DRIVE
Define efficiency, output, installation space and operating conditions
Review power, thermal and control needs to define the development scope
Define evaluation conditions and performance metrics
FOUR-WHEEL CONTROL / DEVELOPMENT STUDY
Allocate force and moment requests to the wheels, then explore tire constraints and the vehicle response over time. Each stage uses its own inputs and model assumptions.
01 / BASIC ALLOCATION
Longitudinal wheel forces sum to a total force; the left–right difference creates yaw moment. This static example calculates allocations that meet both requests under an assumed ±100 N·m torque limit at each wheel.
Each slider endpoint is the maximum with the other request set to zero; the two maxima cannot be achieved together. Feasibility checks only this example’s wheel-torque limits. Tire grip and actual vehicle capability require additional constraints.
| Wheel | Requested torque | Allocated torque |
|---|---|---|
| FLFront left | 25.0 | 25.0 |
| FRFront right | 75.0 | 75.0 |
| RLRear left | 25.0 | 25.0 |
| RRRear right | 75.0 | 75.0 |
Resultant: Fx = 800 N · Mz = 240 N·m. Minimum wheel-torque headroom 25.0 N·m.
For equal front/rear track widths and wheel radii, each side’s torque is split equally between its front and rear wheels. This minimizes the sum of squared torques among allocations satisfying both requests; it does not imply minimum electrical loss.
Axes are x forward, y left and z up; positive Mz is counterclockwise viewed from above. Wheel-end torque is positive when it produces forward force. Whether negative torque brakes or drives in reverse depends on wheel rotation. This static model neglects wheel angular acceleration and rolling resistance, so Fx,i = Ti/r. Tire friction, lateral forces, load transfer and speed-dependent motor and battery constraints belong in the vehicle model.
The concept would coordinate four GaN inverters through a VCU, with each inverter running its current loop locally. It would use an NVIDIA GPU for vehicle-model computation and allocation research, with torque commands and operating status exchanged wirelessly. Torque-command timing and communication-fault behavior are separate design tasks.
03 / VEHICLE RESPONSE
Compare target and actual yaw rates and the vehicle path under a fixed steering input. Change speed, steering and requested drive force, then explore how wheel friction, torque limits and drive faults affect the time response.
Preparing the vehicle yaw response.
We review inverter requirements for Formula-style electric racing (Student E-Formula), automated logistics and robotics, based on the motor and operating conditions of each application.
01E-MOTORSPORT
Drive systems for student electric race cars.
Discuss inverter requirements and technical collaboration for student electric racing projects, based on the power supply, motor and operating conditions.
Discuss an application
02AUTONOMOUS MOBILITY
Traction drives for automated logistics.
Assess inverter power and control requirements against the vehicle’s traction load and operating conditions.
Discuss an application
03ROBOTIC MOTION
Motor drives for robotic systems.
Review inverter requirements and joint development needs for the robot’s motors and control system.
Discuss an application
A PowerCraft project.
Focused on GaN inverters.
PowerCraftMotion is PowerCraft’s GaN HEMT inverter development project. We define application requirements for the supply, motor and operating conditions, then identify the technical collaboration and joint development work required.
Share your motor, supply and operating requirements so we can review the application scope.
02PartnershipDiscuss development objectives, technical scope and responsibilities.
03InvestmentAsk about our technology development and commercialization plans.
Application reviews · Joint development · Investment
