EV Drivetrains

June 19, 2026 19 min read Cars & Vehicles

evmotorsinverterspowertrainsilicon-carbideautomotive

Contents

An internal-combustion car carries a startling amount of machinery whose only job is to compensate for the engine’s flaws. The transmission exists because a piston engine produces useful torque only inside a narrow speed band, usually between 1,500 and 6,000 RPM, and cannot start from rest under load. The differential exists because a single crankshaft must drive two wheels that turn at different speeds in corners. The starter motor exists because the engine cannot start itself. The flywheel, dual-mass damper, torque converter, clutch packs, synchros, and shift forks all exist to paper over the same fundamental problem: combustion produces lumpy, narrow-band torque that needs translation before a wheel can use it.

An electric drivetrain throws nearly all of this away. The motor produces peak torque at zero RPM, runs cleanly to 18,000 RPM or higher, costs almost nothing to keep idling at a stoplight (because it does not idle), and reverses direction by flipping the order of two phase wires in software. The result is that an EV powertrain collapses to three components stacked end to end: an inverter, a motor, and a single-speed reduction gear. The differential is still there, but it is a small open gear-set inside the same housing. There is no clutch, no torque converter, no shift logic, no idle-air-control valve, no starter. A modern dual-motor EV puts one of these units on each axle and gets all-wheel drive without a transfer case or a propshaft. This post walks the three boxes in order, explains why the silicon-carbide inverter changed the math for everyone, and shows where the engineering trade-offs still live, mostly inside the motor itself.

Why a flat torque curve killed the transmission

The case for a multi-speed transmission in an ICE car is almost entirely about staying inside the engine’s useful band. A 2.0-liter gasoline engine might make peak torque of 300 Nm somewhere around 2,500 RPM and peak power somewhere around 5,500 RPM, with usable output collapsing below 1,200 RPM and above 6,500 RPM. To accelerate from a stop to 200 km/h while keeping the engine inside that band, you need a gear ratio that varies by roughly a factor of six. That is what a six-speed gearbox does. Even then the first ratio cannot reach the engine, which is why a clutch or torque converter has to slip on every launch, dissipating energy as heat.

An electric motor’s torque curve is a different shape entirely. From 0 RPM up to a base speed (typically 4,000 to 6,000 RPM), torque is constant at the peak value, limited by current and magnetic saturation. Above base speed the controller enters the field-weakening region, where torque falls roughly as 1/n but power stays approximately constant up to a maximum speed of 15,000 to 20,000 RPM. The ratio between maximum motor speed and “useful from a stop” is therefore on the order of 20:1, more than three times what an ICE delivers. A single fixed gear ratio, usually somewhere between 8:1 and 10:1, lets the motor spin freely all the way to highway speeds while still providing massive wheel torque from rest.

The exceptions are rare and instructive. The Porsche Taycan uses a two-speed gearbox on the rear axle, with a tall second gear that improves high-speed efficiency and top speed. The Audi e-tron GT shares the unit. A few high-end sports EVs from Rimac, Koenigsegg, and the SLS AMG Electric Drive program used multi-speed boxes for the same reasons. But for the vast majority of consumer EVs, from the Tesla Model Y to the Hyundai Ioniq 6 to the BYD Seal, one fixed ratio does everything. The motor is what changes character, not the gearbox.

ICE torque curve              EV torque curve
(peak ~300 Nm @ 2500 RPM)     (peak ~400 Nm flat 0-5000 RPM)

  Nm                            Nm
  |   ____                      |________
  |  /    \___                  |        \___
  |_/         \                 |            \___
  |/           \                |                \___
  +---|----|----|---- RPM       +-----|------|-------- RPM
  1k  2.5k 5k   6.5k             0   5k    15k

  needs 6 gears to use          needs 1 gear, ever

The cascading consequences are large. No transmission means no transmission fluid service. No clutch means no engagement wear. No torque converter means no slip losses at light load. No starter means no flywheel ring gear. The mechanical bill of materials for the propulsion system drops by an order of magnitude, and the failure modes drop along with it. (Our field guide on automatic transmissions walks the planetary, clutch-pack, and valve-body machinery that an EV simply does not need.)

The three motor families and one outlier

Pop the lid on a modern EV drive unit and you will find one of four motor topologies. The choice is driven by a triangle of efficiency, raw-material cost, and rare-earth supply exposure.

Motor type	Peak efficiency	Continuous-rated efficiency	Rare earths	Typical use	Examples
PMSM (interior permanent magnet)	96-97%	92-94%	Yes (Nd, Dy, Tb)	Default for almost everything	Tesla rear unit, Hyundai E-GMP, Kia EV6, Ford Mach-E, GM Ultium, BYD, Lucid
Induction (AC asynchronous)	93-95%	88-91%	None	Front of dual-motor performance EVs	Tesla Model S/X front, Model 3/Y Long Range front
EESM (electrically excited synchronous)	95-96%	91-93%	None	When magnet supply is a strategic worry	BMW i4 / iX / i5, Renault Zoe, some ZF units
SynRM / PMSynRM hybrid	94-96%	90-92%	Reduced	Cost-sensitive new designs	Tesla Model 3 Highland rear (reduced magnet content), some BYD variants

The dominant choice is the interior permanent-magnet synchronous motor (IPM PMSM). The rotor carries embedded neodymium-iron-boron magnets, often with a few percent dysprosium or terbium for high-temperature coercivity. There is no rotor winding, no rotor current, and therefore essentially no rotor copper loss. The stator carries a three-phase distributed winding driven by a variable-frequency inverter, and field-oriented control keeps the stator current vector exactly perpendicular to the rotor flux in the constant-torque region. Peak efficiencies of 96-97% are routine, and the part-load efficiency map stays above 90% across most of normal driving. This is why a PMSM is on the rear axle of nearly every consumer EV sold today. (For the deeper electromagnetic story of BLDC and PMSM control, our steppers, servos, and BLDC primer covers field-oriented control from the bottom.)

The cost of a PMSM is the rotor’s bill of materials. Neodymium and dysprosium are both supply-concentrated in China, both subject to export-license shifts, and both expensive enough that a 200 kW motor can contain $200 to $400 of magnet alone. That is why induction motors survived. Tesla, famously, started with induction motors in the Model S and still uses one on the front axle of dual-motor Model 3, Y, S, and X. An induction rotor is a stack of laminated steel with copper or aluminum bars shorted at the ends; it carries no magnets and no permanent flux. The stator induces rotor currents at slip frequency, and the resulting torque tracks the slip. Peak efficiency is lower because the rotor now dissipates copper and skin-effect losses, but two things make induction attractive on a dual-motor car. First, when the motor is not being driven hard, you can simply zero the stator current and the rotor freewheels with no drag, whereas a PMSM is always cutting through its own magnetic field and generates back-EMF and iron losses even with zero torque commanded. Second, induction handles high overload and high temperature gracefully, which is useful for the occasional-use front axle in a rear-biased performance car.

The electrically excited synchronous motor (EESM) is BMW’s answer to the magnet supply problem. The rotor carries a field winding fed through slip rings or a rotary transformer, and the field current is a third control variable on top of the inverter’s two-axis stator current. EESM efficiency at part load can actually beat a PMSM because the field current can be reduced when torque demand is low, eliminating the constant magnet-flux loss. The cost is mechanical complexity (brushes or contactless excitation) and slightly larger overall package size. BMW shipped EESMs in the i4 and iX and has expanded the family to the i5 and iX1, and several Tier-1 suppliers including ZF and Mahle have catalog EESM units for OEMs who want to avoid magnet exposure.

The synchronous reluctance motor (SynRM) and permanent-magnet-assisted synchronous reluctance (PMSynRM) hybrid sit in between. The rotor is shaped with flux-barrier slots that create magnetic anisotropy; torque comes from the rotor’s preference to align its low-reluctance axis with the stator field. Pure SynRM is rare in cars because torque density is modest, but PMSynRM, which adds a small amount of magnet to bias the reluctance torque, has become a quiet trend. Tesla’s Model 3 Highland rear unit, the 3D6, is widely reported to have reduced its rare-earth magnet content by roughly 25% versus the prior 3D5 unit by going further down the PMSynRM path while keeping power and efficiency essentially the same.

The Tesla dual-motor pairing is a good example of using each motor family for what it does best. The rear PMSM does the heavy lifting and provides high efficiency at the cruise operating point. The front induction motor sits mostly disengaged during light cruising and contributes torque only on launch, hard acceleration, or low-grip events. The result is a car that gets the launch numbers of a dual PMSM with the cruise efficiency closer to a single rear PMSM.

The inverter: where the silicon carbide revolution actually happened

The motor gets most of the marketing attention, but the inverter is where the EV power electronics story has moved fastest in the last five years. The inverter’s job is conceptually simple: take DC from the battery and synthesize three sinusoidal AC currents into the motor’s stator windings, at whatever frequency and amplitude the controller demands. In practice this is a three-phase, two-level voltage-source inverter built from six high-power switches arranged as three half-bridges, switched at a carrier frequency of 8 to 20 kHz using space-vector pulse-width modulation. A current sensor on each phase feeds a Park transform into a d-q reference frame, two PI loops regulate the d and q current components, and an inverse transform turns the desired voltage vector back into PWM duty cycles. Underneath the marketing-friendly term “field-oriented control” sits one of the cleanest applications of PID control from first principles you will ever find.

What changed in the last decade is the switch itself. Through 2017 the standard high-power switch was the silicon insulated-gate bipolar transistor (Si IGBT), a four-layer device that handles 600 to 1,200 V and hundreds of amps but switches slowly, conducts with a fixed 1-2 V offset (the bipolar saturation voltage), and saturates thermally at around 150 C junction temperature. Its loss profile is roughly half conduction and half switching at typical EV operating points, with combined drivetrain inverter losses around 3-4%.

Then Tesla shipped the Model 3 in 2017 with a silicon-carbide (SiC) MOSFET inverter built around discrete 1-in-1 modules from STMicroelectronics. SiC has a wider bandgap than silicon (3.3 eV versus 1.1 eV), which translates to ten times the breakdown field, three times the thermal conductivity, and operating junction temperatures up to 200 C. SiC MOSFETs switch in tens of nanoseconds versus hundreds for silicon IGBTs, conduct with a resistive (not bipolar) drop, and lose far less energy per switching transition. The net effect on a real EV is 1 to 3 percentage points of drivetrain efficiency, depending on duty cycle, which translates to 5 to 10% more range at highway speeds where switching losses dominate. SiC also lets you raise the switching frequency, which reduces motor current ripple and lets the motor designer use less iron and less copper in the stator.

Battery pack (400 V or 800 V DC)
        |
        | high-voltage DC bus
        v
+---------------------+
|  SiC MOSFET inverter|  three half-bridges, 6 switches
|  PWM @ 10-20 kHz    |  space-vector modulation, FOC
+---------------------+
        |
        | three-phase AC (variable f and V)
        v
+---------------------+
|  PMSM / IM / EESM   |  peak 18-20k RPM
+---------------------+
        |
        | mechanical shaft
        v
+---------------------+
|  Single-speed       |  8-10:1 ratio
|  reduction + open   |  helical-cut, oil-cooled
|  differential       |
+---------------------+
        |
        v
       wheels

The other reason SiC matters is that it unlocked 800 V architectures. At 400 V, an IGBT inverter is fine; the switching losses scale with voltage and the IGBT’s bipolar conduction drop is already 1-2 V, so doubling the DC bus to 800 V would roughly double the switching losses and crush efficiency. With SiC the switching loss penalty for 800 V is small enough that the system trade goes the other way: doubling the bus voltage halves the current for the same power, which lets you use thinner copper in the cables and the stator windings, lighter busbars, smaller connectors, and lower-current contactors. Charging benefits proportionally; 800 V cars can pull 250-350 kW from a DC fast charger without melting cables. Porsche Taycan, Hyundai E-GMP (Ioniq 5, Ioniq 6, EV6, EV9), Lucid Air and Gravity, Kia EV6 and EV9, Audi e-tron GT, and the GMC Hummer EV all use 800 V (or near-800 V) buses, all running SiC inverters. The pairing is not a coincidence.

Inverter packaging has followed. The early Tesla SiC inverter used 24 discrete 1-in-1 modules on a pin-fin heat sink, each module containing two SiC dice in parallel. Wolfspeed, ON Semiconductor, Infineon, Rohm, and STMicroelectronics now ship integrated half-bridge SiC modules rated 1,200 V at hundreds of amps, with built-in temperature sensors and gate-driver-friendly Kelvin source pins. Power densities have crossed 50 kW per liter of inverter volume, which is part of how the rest of the drive unit gets so small.

The reduction gearbox: small, simple, oil-cooled

After the inverter and the motor, what is left is the gearbox. In a typical front-wheel-drive EV, the motor sits transversely and its output pinion meshes with an intermediate gear, which drives the final-drive gear on the open differential carrier. Total ratio is between 7:1 and 11:1. The gears are helical-cut steel, ground to high precision because the motor’s high speed means tooth-mesh frequency lives in the human-audible band and any error becomes a whine you cannot ignore.

The lubricant is the unsung hero. Modern EV drive units share a single oil sump between the gear set, the motor bearings, and (in the highest-performance units) the motor windings themselves. The oil is a low-viscosity synthetic, typically around ISO VG 32 or thinner, formulated to be dielectric (because it touches copper windings at full bus voltage), low-foaming, and compatible with copper and rare-earth magnets. Service is generally lifetime or 240,000 km, not the 60,000 km of an automatic transmission fluid.

The really aggressive packaging trick is direct oil-spray cooling of the stator. Lucid, Tesla, Porsche, and several Chinese makers route oil under pressure through jets aimed at the end-turns of the stator winding, where heat is generated by copper losses. Compared to the older approach of pumping water-glycol through a jacket cast into the outside of the stator housing, direct oil cooling cuts the thermal path from the heat source to the coolant from about 8-10 K per kW to under 3 K per kW. That is why continuous power ratings have climbed faster than peak power. The Lucid Air’s drive unit is the headline number here: 670 hp (500 kW peak), motor weight 74 lb (34 kg), entire drive unit including motor, inverter, gearbox, and differential at 163 lb (74 kg). That is roughly 6.8 kW per kg at the drive-unit level, and the unit physically fits in a carry-on suitcase. Tesla’s 3D6 rear unit, Hyundai’s 160 kW E-GMP rear, and BYD’s 8-in-1 unit (which fuses motor, inverter, gearbox, OBC, DC-DC, PDU, BMS, and VCU into one box) are all in the same general family.

A representative spec sheet across the industry, mid-2026:

Drive unit	Peak power	Peak torque	Voltage	Inverter	Motor type	Notes
Tesla 3D6 (Model 3 Highland rear)	208 kW	420 Nm	400 V	SiC	PMSynRM (low-magnet)	replaces 3D5, ~25% less rare earth
Tesla front induction (S/X/3LR/YLR)	137-186 kW	240-310 Nm	400 V	Si IGBT or SiC	Induction	freewheels at cruise
Lucid Air rear unit	500 kW (670 hp)	793 Nm	924 V	SiC	PMSM	74 kg complete
Hyundai E-GMP front	70 kW	255 Nm	800 V	SiC	PMSM	hairpin winding
Hyundai E-GMP rear	160 kW	350 Nm	800 V	SiC	PMSM	2-speed not used
Porsche Taycan rear (Turbo)	335 kW	610 Nm	800 V	SiC	PMSM	2-speed gearbox
BMW iX M70 rear	360 kW	750 Nm	400 V	Si + SiC	EESM	no magnets
BYD 8-in-1 (Seal)	230 kW	360 Nm	800 V	SiC	PMSM	one-box integration
GM Ultium (Hummer EV tri)	3x 250 kW	1,500+ Nm	400 V	SiC	PMSM	three identical units

The numbers cluster more tightly than the marketing would suggest. Almost everyone is now in the 200-500 kW peak range for a single drive unit, 90-95% efficient on the cycle, oil-cooled, and SiC-driven. The Lucid Air’s power density remains an outlier, but it is the kind of outlier that the rest of the industry will narrow toward over the next five years.

Regen blending and one-pedal driving

When the driver lifts off the accelerator (or presses the brake pedal lightly), the inverter does not stop sending current; it reverses the sign of the q-axis current command. The motor now acts as a generator, the inverter as a rectifier with active switches, and energy flows backward from the wheels through the gear set, into the motor’s back-EMF, through the inverter’s body diodes and synchronous-rectification SiC switches, and out into the DC bus. The battery (or, if the battery is full, a chopper resistor) absorbs it.

The catch is that regen torque is not available everywhere on the operating map. Three regions misbehave. Near zero RPM, the back-EMF voltage falls below what the inverter can usefully boost, so regen torque falls off below about 5-10 km/h and disappears entirely at a stop. At very high RPM, the motor is in field-weakening and the achievable braking torque is limited by what the d-axis current can suppress without exceeding inverter current limits. During ABS or stability-control events, the wheels need individually modulated braking torque on a millisecond timescale, which the motor on an axle cannot deliver to one wheel without slipping the differential.

This is why every modern EV has a regen-blending brake controller. The brake pedal is no longer mechanically connected to the master cylinder; instead, a pedal-position sensor and a brake-by-wire actuator (typically a Bosch iBooster or Continental MK C1) combine driver demand with vehicle dynamics. The controller commands as much motor regen as the operating point allows, fills the deficit with hydraulic friction braking, and seamlessly hands off between the two as speed approaches zero. Done well, the driver feels a single consistent deceleration curve. Done badly, you feel the regen drop out at 8 km/h and the hydraulic brakes lurch in to catch you.

“One-pedal driving” is the marketing name for tuning the regen-on-lift map aggressively enough that the driver almost never needs the brake pedal in city traffic. Nissan Leaf e-Pedal, Tesla’s Hold mode, and Hyundai’s i-Pedal all do the same thing: blend in enough hydraulic friction at the very end to bring the car to a true stop and hold it there. The energy recovery rate during urban driving with aggressive regen is typically 60-70% of the kinetic energy that would otherwise have gone into brake-rotor heat. (Why the rest goes elsewhere, and what it does to the battery, is the subject of our piece on EV thermal management; the chemistry that absorbs it sits in our lithium-ion piece.)

Dual-motor AWD without a transfer case

The last piece worth noting is what an EV does for all-wheel drive. In an ICE car, putting power to both axles requires a propshaft running the length of the car, a transfer case to split torque, a center differential or clutch pack to handle the front-rear speed difference in turns, and another driveshaft to the second axle. The mechanical bill is large, the friction losses are noticeable (3-5% on the cycle), and torque distribution between axles is constrained by whatever clutch or viscous coupling the engineers can fit.

A dual-motor EV simply puts one drive unit on each axle, with no mechanical connection between them at all. Front-rear torque distribution is a software variable updated at the inverter control loop rate, typically 10 kHz. You can run pure rear-drive at cruise to keep the front motor’s losses out of the equation, then ramp in 0-50% front torque in milliseconds when the rear slips, when the driver demands launch acceleration, or when stability control wants to reduce yaw. Tri-motor and quad-motor cars (GMC Hummer EV, Rivian R1T, Lucid Gravity, Tesla Cybertruck) extend the same idea per-wheel, with true torque vectoring that no mechanical differential can match. Rivian’s quad-motor variant can rotate the truck around its center axis using opposing wheel torques, a maneuver an ICE truck cannot perform at any price.

The simplification compounds elsewhere. No propshaft means more underbody space for battery modules. No transfer case means less unsprung weight at the center of the car. No center differential means no torque-bias preload, no clutch wear, no fluid service. The mechanical AWD penalty of an ICE drivetrain (extra weight, extra friction, extra failure modes) becomes a small software cost in an EV.

Verdict

An EV drivetrain is not a converted ICE drivetrain with the engine swapped. It is a different topology that exists because the motor’s torque curve made most of the ICE supporting cast obsolete. Three boxes, end to end: a SiC inverter doing space-vector PWM at 10-20 kHz, a PMSM (or induction, or EESM) producing flat torque from zero to base speed and constant power above, and a single-speed helical reduction at 8-10:1. Total mass for a 250 kW drive unit is now under 100 kg, peak efficiency is in the mid-90s, and the failure modes are bearings, gear noise, and inverter capacitors rather than the long tail of clutch, valve-body, and transmission ailments that haunt ICE cars after 200,000 km.

The two open questions are magnet supply and continuous power. The first is being fought on three fronts: lower-magnet PMSynRM designs (Tesla 3D6), wholesale rejection of magnets via EESM (BMW), and aggressive recycling of end-of-life rotors. The second is being chased through direct oil-spray stator cooling and 800 V buses, both of which let you raise the continuous rating without raising peak current. The Lucid Air’s 500 kW unit at 74 kg is the current ceiling; everyone else is moving toward it. The shape of the EV drivetrain has been settled. The remaining engineering is mostly about doing the same job in a smaller box with less neodymium.