Piezoelectric Actuator

What a piezoelectric actuator actually is

Most actuators move things by pushing — a coil pulls iron, a piston pushes fluid, a screw advances on its thread. A piezoelectric actuator does something stranger: it changes shape because its atoms are slightly off-centre and an electric field puts them back. Apply a voltage to a thin slab of ferroelectric ceramic and it physically stretches along the field axis by about a part in a thousand. Remove the voltage and it returns. There is nothing rotating, nothing sliding; the crystal lattice itself is the working mechanism.

That very direct coupling between volts and microns has two consequences. First, the motion is fast — limited only by acoustic propagation in the ceramic, which means microsecond rise times. Second, it is exquisitely fine — there is no minimum step size set by stiction, backlash, friction, or thread pitch. With a well-designed driver you can move a piezo by a single picometre. No other practical actuator family touches that resolution at room temperature.

A 19th-century discovery that waited a century

The piezoelectric effect was discovered in 1880 by Jacques and Pierre Curie, twenty-one and twenty-five years old at the time, working in their Paris lab. They were testing the relationship between pyroelectricity and crystal symmetry and noticed that pressing on certain crystals — tourmaline, quartz, Rochelle salt — produced a measurable surface charge proportional to the applied force. They called the effect piezo, from the Greek for "to press". The following year Gabriel Lippmann, on thermodynamic grounds, predicted the converse effect: that an applied field should produce a strain. The Curies verified it within months.

For sixty years the effect was a laboratory curiosity. Quartz oscillators began to be used in radio in the 1920s — Cady's quartz-controlled frequency standard at Wesleyan in 1921, and later Bell Labs' precision quartz clocks — but the strains involved were tiny and the materials were natural single crystals of limited size. The transformation came in World War II when researchers seeking better sonar transducers discovered barium titanate (BaTiO₃), a ferroelectric ceramic that could be pressed into any shape and then "poled" in a strong DC field to align its domains. Lead zirconate titanate (PZT), discovered at the Tokyo Institute of Technology in 1952, surpassed barium titanate decisively and remains the workhorse of the industry seventy years on.

Material families and their trade-offs

Material	d₃₃ (pm/V)	Curie T (°C)	Typical use	Notes
Quartz (α-SiO₂)	2.3	573	Frequency reference, sensors	Single crystal, very stable, very low loss
Barium titanate (BaTiO₃)	~190	120	Capacitors, simple transducers	Original WWII ferroelectric ceramic
PZT (soft, type 5H)	500-650	193	Actuator stacks, sensors	High d33, high loss, used for displacement
PZT (hard, type 4)	290-320	320	Ultrasonic transducers	Low loss, high Q, withstands high drive
PVDF (polymer)	~30 (d33), 23 (d31)	~80	Flexible sensors, hydrophones	Cheap, flexible, low force
PMN-PT single crystal	2000+	~140	Premium medical transducers	Highest performance, low Curie T, expensive
KNN / lead-free ceramics	250-400	~250	EU RoHS-compliant designs	Lead-free, performance still trailing PZT
AlN (aluminium nitride)	~5	—	MEMS, RF resonators	CMOS-compatible, no lead, low coupling

The d₃₃ coefficient is the central figure of merit — it tells you how much strain per unit applied field, in metres per volt (or equivalently picocoulombs per newton for sensing). A 5 mm-thick slab of soft PZT with d₃₃ = 600 pm/V driven at 1 kV/mm gives 5 mm × 1000 V/mm × 600 × 10⁻¹² m/V = 3 μm of stroke. Doubling d₃₃ doubles the stroke for the same field; multilayering the same slab into 50 thin layers preserves stroke while dropping drive voltage by 50×.

How a ferroelectric crystal turns volts into motion

PZT is a perovskite — chemical formula Pb(Zr,Ti)O₃, with lead ions at the cube corners, oxygen at the face centres, and a titanium or zirconium ion near the centre of each cube. Above the Curie temperature the structure is cubic and centrosymmetric; below it the central ion shifts a few picometres off-centre, producing a permanent electric dipole in every unit cell. The crystal is now ferroelectric.

A virgin ceramic has its dipoles oriented randomly across grains, so the bulk polarization is zero and the piezo response averages out. The crucial manufacturing step is poling: heat the ceramic just below T_c, apply a DC field of 2-4 kV/mm for several minutes, and the dipoles all rotate into the closest allowed direction (the <111> or <001> depending on composition). After cooling, the polarization is locked in. The ceramic now behaves like a single-domain crystal at room temperature.

An applied field along the poling direction does two things. First it slightly shifts the off-centre ion further in the field direction, elongating the unit cell — the linear piezoelectric strain, governed by the d₃₃ coefficient. Second, at higher fields, it can flip individual domains by 90° or 180°. The first mechanism is reversible and roughly linear; the second is hysteretic and is what limits open-loop accuracy.

S = d · E      (strain = piezo coefficient × electric field)
ΔL = d₃₃ · V   (axial displacement for a layer of any thickness)
ΔL_stack = N · d₃₃ · V   (multilayer stack of N layers)

Notice that for a single layer the displacement does not depend on the thickness — it depends only on the voltage. Halving the layer thickness halves the field needed for the same strain and halves the voltage. That is the whole reason multilayer stacks exist.

The multilayer stack — engineering parallel layers in series

A modern piezo stack actuator is a stack of 100 to 1000 thin PZT layers, each about 20-100 μm thick, with interdigitated metal electrodes co-fired into the ceramic during sintering. Adjacent layers are electrically parallel — every odd electrode connects to terminal A, every even electrode to terminal B — but mechanically in series. So when you drive the stack with 150 V across the terminals, every layer sees 150 V across its 50 μm thickness, or 3 kV/mm, and strains by 0.15%. A 200-layer stack of 50 μm layers is 10 mm long, and the 0.15% strain adds to 15 μm of total stroke.

The same 10 mm of bulk PZT driven as a monolithic block would need 30 000 V to develop the same strain — impossibly inconvenient and a sure recipe for dielectric breakdown. Multilayering is the trick that makes piezo actuation practical with battery-and-rail-compatible drive electronics.

Co-firing the electrodes inside the ceramic is itself a tour de force of materials processing. The internal electrodes have to be made of metal that survives the 1000-1300 °C sintering temperature of PZT — early stacks used pure platinum or palladium; modern designs use silver-palladium alloys or even pure copper in reducing atmospheres. Every fired layer must be free of cracks and voids over its full area, because a defect that bridges two electrodes turns into a hot-spot short under field.

Performance metrics — stroke, force, bandwidth

Three numbers characterise an actuator. Free stroke is the displacement at zero load. Blocked force is the force at zero displacement. The operating curve between them is a straight line for an ideal piezo, and the useful mechanical work per cycle is ΔL_free · F_blocked / 4 — maximised when the load stiffness equals the actuator stiffness.

Class	Free stroke	Blocked force	Drive voltage	Bandwidth	Example use
Tube scanner (STM)	~1-10 μm xy, 1 μm z	N to tens of N	±150 V	kHz	STM, AFM head
Multilayer stack (small)	5-15 μm	500-3000 N	100-200 V	~10 kHz	Optics positioning, nanopositioning
Multilayer stack (large)	30-100 μm	5-50 kN	100-200 V	~5 kHz	Fuel injector, ultrasonic horn driver
Bimorph bender	0.1-1 mm	0.1-10 N	100-200 V	~100 Hz - 1 kHz	Inkjet, autofocus, micro-pump
Ultrasonic transducer	~nm at resonance	—	10-1000 V	20 kHz - 50 MHz	Medical imaging, NDT, cleaning
Walking / inchworm	millimetres or more	10-100 N	±150 V	mm/s	Long-range nanopositioning

The bandwidth ceiling is set by the mechanical resonance of the stack plus its load. A 10 mm-tall stack has a free resonance around 50 kHz; once you bolt it to a payload the frequency drops. For continuous AC drive at large amplitude the limit is usually thermal — every cycle dissipates ~10% of the input as heat through hysteresis, and the ceramic depoles if it warms past T_c/2.

Compared to other actuator families

Family	Stroke	Force	Bandwidth	Precision	Strengths / weaknesses
Piezo stack	μm	kN	kHz - MHz	sub-nm	Fast, precise, stiff, but short stroke and hysteretic
Electromagnetic solenoid	mm - cm	N - 100 N	kHz	~μm at best	Cheap, long stroke, moderate force; bulky
Voice coil	mm - cm	N - 10 N	kHz	μm	Linear, smooth, used in HDD heads & speakers
Servo motor + leadscrew	m	kN	~100 Hz	~μm with encoder	Long stroke; backlash, friction, slow inversion
Hydraulic ram	m	MN	~50 Hz	~mm	Huge force; slow, dirty, needs pump
Pneumatic cylinder	m	kN	~10 Hz	poor	Cheap, soft, imprecise
Shape-memory alloy	cm	N - 100 N	Hz	poor	High specific work; very slow, narrow temperature window
MEMS electrostatic	μm	μN - mN	kHz	nm	Tiny; pull-in instability limits stroke to a third of gap

The niche piezo dominates is "small motion, fast, precise, stiff". When you need many millimetres of stroke at moderate precision, electromagnetic is cheaper. When you need megan­ewtons of force, hydraulics still wins. But for nanometre positioning at kilohertz bandwidth — STM tips, optics in lithography, fast steering mirrors, atomic-force microscope cantilevers — no other technology is close.

Where piezo actuators show up

• Scanning probe microscopy. STM and AFM heads use piezo tube or stack scanners to raster their probe over a sample with sub-Angstrom z-resolution. The 1981 IBM Zurich STM that won Binnig and Rohrer the Nobel was built around a hand-fabricated three-axis piezo tripod. Modern instruments use commercial flexure-guided multilayer stacks closed-loop with capacitive sensors. Without piezos, none of atomic-resolution surface science would exist.
• Inkjet print heads. Epson DURABrio, Konica Minolta KM1024, Kyocera KJ4 and Ricoh Gen5 heads use PZT diaphragms or shear-mode plates to eject 1-10 picolitre droplets at 20-100 kHz per nozzle. They tolerate solvent and UV-cured inks that boil-bubble HP-style thermal heads will not survive. Industrial single-pass inkjet printers and 3D-printing material jetters all use piezo heads.
• Diesel piezo fuel injectors. Bosch CRI3 piezo common-rail injectors use a stack of about 250 PZT layers — roughly 30 mm tall, drawing 150 V — to drive the injector needle directly without an intermediate solenoid valve. The result is sub-100 μs switching, allowing up to five injection events per combustion cycle, which is critical for meeting modern NOx and particulate emissions limits. Adopted in the late 2000s by Mercedes (OM642), Audi (3.0 TDI) and BMW (M57/N57) V6 diesels.
• Ultrasonic transducers. Every medical ultrasound transducer, every ultrasonic non-destructive test probe, every ultrasonic cleaner, every dental scaler, every phacoemulsification cataract handpiece is a PZT or PMN-PT resonator. Operated at its mechanical resonance, the same stack that delivers 10 μm DC delivers ±nm at megahertz, and the radiated pressure is enormous.
• Active vibration cancellation. Bolted between a vibrating machine and a sensitive payload, a stiff piezo stack with closed-loop feedback can cancel multi-kilohertz disturbances down to a few nanometres residual. Used on optical tables, satellite reaction-wheel isolators, and high-end semiconductor lithography stages.
• MEMS micromirrors and varifocal lenses. Thin-film PZT or AlN MEMS layers can tilt or deflect silicon mirrors at kilohertz rates, an alternative to Texas Instruments' electrostatic DMD architecture for laser projection, LiDAR scanning, and confocal microscopy. PiezoMEMS varifocal lenses replace voice-coil autofocus in tiny imaging systems.
• Smartphone haptic feedback. Apple's Taptic Engine and competing Android haptics use a tiny piezo or linear-resonant actuator to produce the crisp tactile clicks of modern touchscreens. The keys on virtual keyboards "feel" because a 50 μm piezo wafer pulses against your fingertip at the moment of contact.
• Camera optical image stabilisation. High-end cameras and increasingly phones use piezo benders or ultrasonic motors to shift sensor or lens elements at hundreds of hertz, cancelling hand tremor with sub-pixel precision.

Hysteresis, creep, and how to control a piezo

A piezo is not a clean spring; it has three intrinsic nonlinearities that complicate precision control.

• Hysteresis. The strain-versus-voltage curve traces out a loop; rising voltage takes one path, falling voltage another. Loop width is typically 10-15% of full stroke. The classical Preisach model represents this as a superposition of elementary hysterons; modern controllers fit a Bouc-Wen or Prandtl-Ishlinskii model and apply its inverse as feedforward.
• Creep. After a voltage step the strain continues to drift logarithmically for minutes — typically 1-2% of the step over the first hour. Compensate either with closed-loop position feedback or with empirical creep curves applied as a slow correction.
• Thermal drift. d₃₃ varies with temperature, and the actuator self-heats under AC drive. For nanometre-class instruments the actuator and its sensor must share the same thermal mass and the system is allowed to equilibrate before measurement.

The standard control architecture is closed-loop: a capacitive, eddy-current, or strain-gauge sensor reads the actual displacement to picometres; a digital PID-plus-feedforward controller compares it to the command and drives the high-voltage amplifier. Step settling to within 1 nm in 1 ms is routine in commercial nanopositioners from Physik Instrumente (PI), nPoint, and Mad City Labs.

Mounting — flexures, preload, and why you must never bend a stack

PZT is a ceramic. It is strong in compression (around 250 MPa) but extremely weak in tension or bending (around 20 MPa); a transverse load that the bare ceramic would shatter under in milliseconds is routinely encountered in any mechanism that includes a piezo. The conventional fix is twofold: preload the stack with a compression spring delivering 10-20% of its blocked force, so the ceramic remains in net compression under all working conditions; and guide the motion with a flexure rather than allow it to bear the load through sliding bearings. A flexure-guided piezo stage uses a monolithic stainless-steel or aluminium block machined into thin-walled hinges that constrain the motion to a single axis while transferring zero transverse load to the ceramic.

This pairing — piezo stack plus parallel-leaf flexure — is the foundation of every commercial nanopositioner. The flexure also provides the restoring force when the drive voltage drops, and integrates cleanly with capacitive position sensors that read across the flexure gap.

Driving a piezo — the unappreciated power electronics

A piezo stack looks electrically like a several-microfarad capacitor with a small loss tangent and a strong dependence of capacitance on signal amplitude. Driving it at full stroke and kHz bandwidth requires a high-voltage amplifier that can source and sink substantial current — typically amperes — because the stack must be both charged and discharged on every cycle. A 1 μF stack swung to 200 V at 1 kHz absorbs about 0.4 W average and demands a peak current of 1.3 A.

For low-duty-cycle DC positioning a linear high-voltage amplifier suffices. For high-bandwidth AC operation, modern drivers use switch-mode regenerative topologies that recover the reactive energy on the falling edge of each cycle, returning it to the supply. Otherwise the power dissipation becomes prohibitive: a continuously-operating large piezo at full bandwidth could pull a kilowatt that almost all has to go somewhere.

Common pitfalls

• Operating the stack in tension. Ceramic in tension fails brittlely. Always preload. Even a momentary inertia-driven reversal can crack a stack; the preload must exceed the worst-case dynamic transverse and tensile load.
• Exceeding the Curie temperature. Self-heating under continuous AC drive can depole the ceramic permanently. Above roughly half T_c the response degrades; above T_c it is lost. Datasheet duty cycle limits exist for a reason.
• Open-loop nanopositioning expectations. Without a sensor in the loop, hysteresis and creep set the floor at about 1-2% of full stroke — perhaps 100 nm on a 10 μm stack. Sub-nm positioning needs a sensor.
• Ignoring the capacitive load. Stack manufacturers quote bandwidth into a "moderate" load. Plugging the stack into a generic op-amp output that cannot source the reactive current results in distorted waveforms and unexpected resonance peaking.
• Forgetting that strain coefficients drift. Long-term aging of poled PZT reduces d₃₃ by ~1% per decade of time after poling. Logarithmic, so most of it is in the first months, but a one-year-old actuator is not identical to a one-day-old one.
• Confusing d₃₃ with d₃₁. d₃₃ is the axial coefficient (strain along the field); d₃₁ is the transverse coefficient (strain perpendicular to the field) and is typically negative and roughly half the magnitude. Bender and bimorph designs exploit the d₃₁ mode; stack designs use d₃₃.