Choose air bearings when you need friction-free motion and nanometer smoothness over long travel. Choose flexures when travel is short and zero backlash matters most. Choose crossed roller bearings when you need high stiffness and load capacity at moderate cost. The wrong pick corrupts the error budget before the first part ships.

How Air Bearings Work

An aerostatic bearing never lets two solid surfaces touch. Clean, filtered air at roughly 4 to 6 bar feeds through restrictors into the gap between the bearing and its guide surface. The pressurized film, typically 5 to 20 microns thick, floats the moving carriage with effectively zero static friction. That is the entire trick: a fluid film replaces mechanical contact, so motion is smooth down to the noise floor of the guide geometry itself.

Two restrictor designs dominate air bearing design. Orifice restrictors meter flow through small holes drilled into the bearing face. They are cheap and easy to manufacture, but each orifice creates a localized pressure peak with a low-pressure zone around it, which produces an uneven stiffness map across the bearing face. Worse, an orifice and the air film can couple into an oscillator. That instability is called pneumatic hammer, and it produces self-excited vibration that no controller can remove because it lives inside the bearing itself.

Porous media air bearings solve this by replacing drilled orifices with a permeable material, usually graphite or a sintered metal, that distributes airflow evenly across the whole bearing face. The pressure field becomes uniform, the effective load area grows, and the hammer instability is strongly suppressed because there is no discrete orifice volume to pump. Porous bearings tolerate minor surface imperfections better and typically deliver 10 to 30 percent more stiffness than an equivalent orifice design. Their weakness is contamination sensitivity: a clogged pore is permanent damage, so the air supply needs sub-micron filtration and the working surfaces must stay clean.

Anatomy of an air bearing is straightforward: a rigid housing, the bearing face with its restrictor system, the air gap, and the guide surface. What people miss is the air bearing system around it. The bearing is only as good as its supply. You need a compressor, dryers, coalescing filters, pressure regulators, and often a dedicated clean circuit. Pressure drop anywhere in that chain directly reduces air bearing stiffness, because stiffness scales with supply pressure. Design the plumbing with the same care as the bearing faces, or the expensive bearing never performs.

For deeper background on the technology itself, see the overview of air bearings.

How Flexure Bearings Work

A flexure bearing moves by elastic deformation, not by sliding. Thin sections of metal bend within their elastic range to guide a stage along one or more degrees of freedom while remaining extremely stiff in every other direction. Because nothing slides and nothing rolls, a flexure has zero friction, zero backlash, zero wear, and zero hysteresis in the mechanism itself. Repeatability is limited only by the actuator and the sensor, not by the bearing. That is why flexures dominate nanopositioning stages, optical alignment mechanisms, and short-stroke fine positioning axes.

The design rules are unforgiving. Stress in the flexure blades must stay below the material’s fatigue endurance limit, or the bearing cracks after some number of cycles. Good practice holds peak bending stress under about one third of yield for high-cycle applications, with margins verified by finite element analysis and fatigue testing. Material choice matters: spring steels, beryllium copper, titanium alloys, and precipitation-hardening stainless grades each offer different combinations of strength, stiffness, and thermal behavior. Hinges are usually cut by wire EDM for clean stress distribution, since sharp corners act as crack initiation sites.

Travel range is the hard limit. A flexure blade can deflect only a small fraction of its length before stress becomes excessive, so flexure bearing travel typically ranges from hundreds of microns to a few tens of millimeters. Long travel demands long blades, which drop the constrained-direction stiffness and invite buckling and parasitic resonances. Monolithic flexures also couple motions: push a parallel-blade stage and the moving platform arcs slightly instead of traveling perfectly straight. That parasitic motion must be modeled or compensated, which adds design work that novices underestimate.

How Crossed Roller Bearings Work

A crossed roller bearing uses cylindrical rollers arranged at right angles to each other between two V-shaped raceways. Each roller contacts the raceway along a line rather than a point, so the contact area is large and the assembly carries high loads in every direction: radial, axial, and moment. The crossed arrangement means a single compact bearing handles combined loading that would need multiple angular-contact ball bearings otherwise.

The defining design variable is preload. Manufacturers set a light, standard, or heavy preload by selecting roller diameter and raceway geometry during assembly. Preload squeezes out internal clearance, which raises stiffness dramatically and eliminates play. But preload is also the bearing’s failure path. Too much preload raises friction and shortens life; too little leaves clearance that shows up as lost motion. Worse, preload relaxes over time through roller and raceway wear, thermal cycling, and settlement of the mating structure. A machine that holds tolerance on day one can develop measurable backlash after two years of production.

Crossed rollers are the pragmatic middle ground of precision motion. They need no air supply, tolerate factory environments, and deliver high stiffness per dollar. Their error motion is larger than an air bearing’s because roller diameter variation, raceway waviness, and cage effects all imprint on the output motion. In a rotary table that shows up as synchronous error once per revolution; in a linear guide it appears as straightness variation along travel.

Head-to-Head Comparison

No single bearing wins every axis. Rank the three technologies against the dimensions that actually drive selection decisions:

Two patterns jump out. Air bearings trade stiffness and robustness for unmatched smoothness and zero wear. Flexures trade travel for perfect repeatability. Crossed rollers trade ultimate smoothness for stiffness, load capacity, and practicality. Every selection argument flows from those trades.

Stiffness and Damping: The Honest Numbers

Bearing catalogs advertise peak stiffness. Machines live at system stiffness. Keep these honest ranges in mind when building an error budget.

Air bearing stiffness is the softest of the three and the most conditional. A precision linear air bearing slide typically shows 20 to 80 N per micron of lateral stiffness at 5 to 6 bar supply pressure. Rotary air bearing spindles run stiffer radially, commonly 50 to 250 N per micron, because the full circumference carries load. Cut the supply pressure in half and stiffness falls roughly with it. Let the gap grow from 10 to 15 microns through wear or poor setup and stiffness drops further, since air film stiffness is strongly gap dependent. Quote stiffness numbers only with the pressure, gap, and temperature stated, or they are meaningless.

Damping is where air bearings disappoint newcomers. The damping ratio of an aerostatic film is commonly 0.01 to 0.05, meaning a shock or a servo disturbance rings for many cycles. Squeeze-film damping helps at high frequency, but it does not rescue a servo loop with poor gain margin. If your machine has a cutting force, a reaction force from a linear motor, or a floor vibration problem, plan for external damping: constrained-layer damping in the structure, tuned mass dampers, or active vibration isolation. Do not discover the damping problem during commissioning.

Flexure stiffness is directional by construction. A well-designed XY flexure stage holds 100 to 500 N per micron in the blocked axes while the motion axis sits at 1 to 10 N per micron by intent. That asymmetry is the feature: the actuator only fights the soft axis. Resonance frequencies of the constrained modes sit high, often 500 Hz to several kHz, which keeps them out of the servo bandwidth. Damping in metal flexures is low, so stages ring at their natural frequencies; add damping treatments or shape the command profile to avoid exciting them.

Crossed roller bearings are the stiffest per dollar. A mid-size linear crossed roller guide with standard preload delivers 200 to 600 N per micron. Large rotary crossed roller bearings for machine tool tables exceed 1000 N per micron. The rolling contacts plus grease films give these bearings the best passive damping of the three, which is one reason machine tools tolerate their error motion: the structure stays quiet. Remember that catalog stiffness assumes correct mounting on a stiff, flat structure. A soft mounting plate or an uneven bolt pattern can cut realized stiffness in half.

When Air Bearings Win

Choose air bearings when the error budget is dominated by motion smoothness and friction, not by static stiffness. The classic cases:

Metrology spindles and scanning axes. Roundness testers, surface profilometers, and wafer inspection stages need motion with nanometer-level smoothness and no stick-slip. Rolling element error motion imprints directly on the measurement. Air bearings win because the fluid film averages out guide imperfections and contributes no friction signature of its own.

High-speed precision spindles. Grinding, micromilling, and diamond turning at tens of thousands of rpm punish any contacting bearing with heat and wear. Air spindles run cool and wear-free for years at speeds that destroy ball bearings. Pair the spindle with proper spindle metrology practice so you can verify error motion instead of assuming it.

Cleanroom and vacuum-compatible axes. Air bearings shed no particles and need no grease, which suits semiconductor and optics environments. Vacuum applications need special designs (differential pumping around the exhaust), but the no-lubricant property is a genuine advantage where contamination control dominates.

Long-travel precision stages. When a stage must travel 300 mm with sub-micron straightness, a linear air bearing on a granite beam beats a flexure (impossible at that travel) and beats crossed rollers on smoothness, provided the machine can carry the air supply and the stiffness is adequate for the process forces.

Air bearings lose when process forces are large, when the environment is dirty, or when nobody will maintain the air system. A grinding operation with 200 N of cutting force on a 40 N per micron air slide deflects 5 microns. That may exceed the tolerance. Be honest about the force levels before committing.

When Flexures Win

Choose flexures when travel is short, repeatability is the requirement, and the environment is hostile. The classic cases:

Nanopositioning and fine alignment. Piezo-driven flexure stages routinely repeat to single nanometers because nothing in the mechanism slips, sticks, or wears. Optical fiber alignment, lithography focus stages, and scanning probe microscope scanners all rely on this property.

Harsh or maintenance-free environments. Vacuum chambers, cryogenic setups, radiation areas, and remote installations punish bearings that need lubrication or air. A monolithic flexure has nothing to service, nothing to leak, and nothing to contaminate. Design the fatigue life once and it runs for decades.

Fast small oscillations. Flexures excel at dither mirrors, choppers, and vibration compensation actuators where the motion is small, fast, and continuous. The low moving mass and high constrained-direction stiffness push resonances out of the operating band.

Flexures lose whenever travel exceeds their elastic range. Do not try to stretch a flexure to 100 mm of travel; the blades get long, the constrained stiffness collapses, and buckling becomes the failure mode. If you need long travel plus fine positioning, use a coarse-fine architecture: a crossed roller or air bearing stage for the long stroke, carrying a flexure stage for the fine axis. This stacked approach is standard in precision machines and costs less than fighting physics.

When Crossed Roller Bearings Win

Choose crossed roller bearings when the machine must carry real loads, hold stiffness under force, and survive a factory floor. The classic cases:

Machine tool axes and rotary tables. Milling, turning, and grinding apply hundreds to thousands of newtons. Crossed rollers deliver the stiffness to hold tolerance under those forces, the damping to keep chatter down, and the load capacity to support heavy workpieces. No other option here survives the cost and robustness argument.

General precision automation. Pick-and-place, dispensing, and assembly machines need good accuracy at moderate cost with long travel. Crossed roller linear guides are the default answer: stiff enough, accurate enough, available in standard sizes, and tolerant of imperfect maintenance.

Moment-loaded structures. Robot wrists, goniometers, and tilt tables see large overturning moments. The crossed arrangement resists moments in a compact envelope that would need a much larger air bearing or a complex flexure to match.

Crossed rollers lose when the accuracy requirement drops below about a micron of error motion, when any friction signature is unacceptable, or when maintenance will be neglected. If the application measures nanometers, move up to air or flexure.

Failure Modes That Actually Kill Bearings

Pneumatic hammer. The air bearing killer. A compressible volume behind an orifice restrictor coupled with the spring of the air film forms an oscillator. If the restrictor volume is too large relative to the film stiffness, the bearing vibrates spontaneously at tens to hundreds of hertz, ruining the surface finish or the measurement. Prevention is design discipline: minimize dead volume between restrictor and film, keep the supply pressure stable, and prefer porous media restrictors, which distribute the flow and remove the discrete pumping volume. If a bearing hammers on commissioning, the usual fix is a smaller restrictor or a redesign of the feed geometry, not more pressure.

Air supply contamination. Oil carryover from compressors, water from inadequate drying, and particles from poor filtration clog restrictors and score bearing faces. Porous media bearings are especially vulnerable because clogged pores cannot be cleaned. The failure looks like a slow loss of stiffness and load capacity over months, then a crash when the film collapses locally. Specify coalescing filters to 0.01 micron, a refrigerated or desiccant dryer holding the pressure dew point at least 10 K below ambient, and pressure regulation with a proper receiver. Monitor pressure at the bearing, not just at the compressor.

Flexure fatigue. Flexures do not wear out; they crack. The failure mode is fatigue at stress concentrations: the root of a notch hinge, a wire-EDM recast layer, or a scratch from handling. Prevention starts with honest stress analysis including stress concentration factors, continues with surface finish control (polish or shot-peen critical hinges), and ends with a proof test at over-travel. Never exceed the designed travel with a hard stop missing. One over-travel event can start a crack that propagates silently for months before the stage loses accuracy.

Roller brinelling and preload loss. Static overload dents raceways (true brinelling), creating permanent dents that show up as periodic error motion. More common is false brinelling: fretting corrosion at roller contact points during micro-oscillation or vibration in transport, which etches the raceway without any overload. Prevention means sizing the bearing for the real static loads including crashes, using transport locks, and scheduling preload verification as part of preventive maintenance. A bearing that was perfect at installation is not guaranteed perfect at year three.

Integration Notes

Air supply design. Treat the air system as part of the bearing. Size the compressor for continuous duty at the bearing flow rate plus margin, because air bearings consume air constantly, not intermittently. A typical small air spindle uses 30 to 100 liters per minute at 6 bar; a large linear stage can use several hundred. Pipe with smooth bends and generous diameters to avoid pressure drop, and place the final regulator and filter as close to the bearing as possible. Add a pressure interlock that disables motion when supply pressure falls below the safe threshold. A machine that crashes its air bearings during a power dip will remember the lesson expensively.

Preload and mounting. Crossed roller bearings achieve their stiffness only with correct preload and a proper mounting structure. The mounting faces must be flat and stiff; a bearing bolted to a compliant plate behaves like a compliant bearing. Follow the manufacturer’s torque sequence and verify with a dial indicator or, better, measure the realized stiffness under load. For air bearings, the equivalent of preload is the guide geometry: straightness and flatness of the guide surfaces directly set the error motion, so budget for precision grinding or lapping of granite and steel ways.

Thermal behavior. All three bearing types are thermally sensitive, but differently. Air bearings exhaust supply air across the guide surfaces, which can actually stabilize temperature if the air is conditioned, but compressor heat and pressure fluctuations show up as drift. Flexures are usually monolithic, so thermal expansion is uniform and predictable; the risk is thermal gradients from actuators mounted nearby. Crossed rollers generate heat from rolling friction and preload, especially at speed, and the steel structure expands accordingly. In every case, put the metrology loop materials and the thermal management on the same design page from day one. A bearing chosen for its error motion spec will not hold that spec in a machine with a 5 K thermal gradient across the frame.

These integration details are where machine design services earn their keep: the bearing datasheet number is the starting point, and the realized performance comes from the system around it.

A Real-World Selection Scenario: The Wafer Inspection Stage

Consider an illustrative scenario drawn from typical precision equipment practice: a wafer inspection stage that must scan a 300 mm wafer under an optical sensor with 100 nm positioning repeatability and a scan speed of 200 mm per second. The error budget allows 50 nm of bearing-induced motion. Three candidate architectures go on the table.

The crossed roller option is the cheapest and stiffest. A precision-grade linear guide set delivers roughly 2 microns of straightness error over 300 mm, an order of magnitude over the budget. Software error mapping could compensate the repeatable portion, but roller error motion includes non-repeatable components from cage movement and lubricant variation. The mapping helps but does not reach 50 nm reliably. Crossed rollers lose on accuracy, not on cost.

The flexure option offers perfect repeatability but cannot cover 300 mm of travel. A stacked design becomes necessary: a coarse stage for the long stroke plus a flexure fine stage for the last few millimeters. This works and is a legitimate architecture, but it adds an actuator, a sensor, and a control loop. Complexity rises, and the coarse stage still needs a bearing, which brings the selection question back one level.

The air bearing option fits the problem directly. A linear air bearing slide on a granite beam travels the full 300 mm with sub-100 nm straightness achievable on good geometry, contributes no friction signature to disturb the servo, and never wears. The costs are real: a conditioned air supply, a clean environment, and lower stiffness that forces attention to reaction forces from the linear motor. In this scenario the stiffness check passes because the process forces are small: the sensor is non-contact, so the dominant disturbances are acceleration forces and floor vibration, both manageable with a stiff granite structure and vibration isolation.

The decision: air bearings for the scan axes, with a flexure-based fine Z axis for focus. The crossed roller option stays in the design as the wafer loading and transfer mechanism, where travel and load capacity matter and nanometer accuracy does not. This kind of mixed architecture is the norm in real machines. Few precision systems use one bearing technology everywhere; the skill lies in assigning each axis the bearing whose strengths match that axis’s error budget. A comparable real project, the wafer transfer module project, shows how bearing and mechanism choices get resolved inside a complete machine design rather than on a datasheet.

Notice what drove the decision. Not the catalog stiffness, not the price of the bearing, but the error budget of the specific axis, the process forces, and the environment. Write those three things down before opening any catalog. The engineers who skip that step end up redesigning the stage in the second year.

FAQ

Do air bearings really have zero friction? Static friction is effectively zero because no solids touch, so there is no stick-slip and no breakaway force. Viscous shear in the air film creates a tiny velocity-dependent drag, but it is orders of magnitude below rolling or sliding friction and entirely predictable, which is why servo loops tune so cleanly on air bearing stages.

Why are porous media air bearings better than orifice types? Porous restrictors distribute airflow across the entire bearing face instead of concentrating it at drilled holes. This produces a uniform pressure field, higher effective stiffness, and far less risk of pneumatic hammer instability. The tradeoff is cost and contamination sensitivity: clogged pores cannot be cleaned, so the air supply needs excellent filtration.

Can a flexure bearing really last forever? It can outlast the machine if the design keeps peak stress below the material’s fatigue endurance limit. Flexures fail by fatigue cracking, not wear, so life is a stress problem, not a mileage problem. Over-travel events, stress concentrations at sharp corners, and poor surface finish are what kill flexures. Design the travel stops first, not last.

When should I avoid crossed roller bearings? Avoid them when the accuracy requirement drops below roughly a micron of error motion, when any friction or stick-slip signature would corrupt the process, or when the machine will run without maintenance. Roller and raceway geometry imprint on the output motion, preload relaxes over years, and neglected lubrication shortens life. Match the bearing to the maintenance reality, not the wish.

Bearing selection is an error-budget exercise, not a catalog exercise. Define the accuracy, forces, travel, environment, and maintenance reality for each axis, then match the technology to those numbers. When the tradeoffs are close or the budget is tight, a structured comparison pays for itself: request a bearing trade study and get the numbers checked before the design freezes.