Deformable mirror

A deformable mirror (DM) is a mirror whose reflective surface can be reshaped under electronic or pneumatic control, so that it changes the phase or focus of the light it reflects. Deformable mirrors are the correcting element in adaptive optics, where a wavefront sensor measures distortions in incoming light and the mirror is reshaped to cancel them, a feedback loop that runs several hundred times a second in astronomical systems. The idea comes from astronomy, where Horace W. Babcock proposed it in 1953 to compensate for atmospheric seeing. It was later carried into vision science, where a deformable mirror corrected the aberrations of the human eye well enough to image single cells in the living retina.

In virtual reality (VR) and augmented reality (AR) research, deformable mirrors, and especially thin deformable membrane mirrors, are one way to build varifocal and multifocal near-eye displays. Changing the curvature of the mirror changes its optical power, which moves the apparent depth of the displayed image. The aim is to supply correct focus cues and reduce the vergence-accommodation conflict of conventional fixed-focus head-mounted displays. Research prototypes include retinal scanning displays built at the University of Washington in the early 2000s, a multi-focal-plane optical see-through display from the University of Arizona, and a wide-field see-through display from the University of North Carolina at Chapel Hill and its collaborators that won the IEEE VR 2017 best paper award.

Reviewed 11 October 2026. Checked every claim against the cited papers and pages (Davies and Kasper 2012, Babcock 1953, Williams 2011, Liang et al. 1997 abstract, Kramida 2016, Pribošek et al. 2023, Dunn et al. 2017 and 2018 full texts, Traub 1967 abstract, ASME DMD brochure, IEEE VR 2017 awards, Road to VR, GeekWire) and bibliographic data via Crossref. About review dates.

How it works

In an adaptive optics system the deformable mirror usually consists of an array of actuators attached to a thin optical surface, which deforms as the actuators expand. Davies and Kasper list stroke (how far the surface can move), response time, actuator spacing and number of actuators as the most important parameters of a DM. Actuator spacing and response time must match how quickly and on what spatial scale the atmospheric distortion changes, while the required stroke and actuator count grow with telescope aperture. They wrote in 2012 that the largest DMs then had some thousand actuators, and that AO at visible wavelengths on a 40 m telescope would need several tens of thousands.

Davies and Kasper describe three main technologies used for astronomical DMs:

Type	Actuation	Characteristics reported by Davies and Kasper (2012)
Adaptive (deformable) secondary mirror	Voice coils, locally positioned by an internal control loop	Actuators a few cm apart behind an optical shell about 1 m in diameter but only 1-2 mm thick; replaces the telescope's secondary mirror, avoiding extra relay optics. The first was installed in 2003 at the 6.5 m MMT.
Piezoelectric DM	Piezo stacks	Actuator spacing of several millimeters; about 10 µm peak-to-valley stroke; response time of the order of a hundred microseconds; affected by hysteresis and thermal expansion, so it must be controlled by the AO system to hold a precise, stable wavefront.
MEMS (MOEMS) DM	Electrostatic or voice-coil, made with semiconductor fabrication	Inter-actuator spacing of a few hundred microns; almost instantaneous response, no hysteresis, relatively cheap with many actuators; some have only about 2 µm of stroke and are paired with a larger-stroke "woofer" mirror.

MEMS deformable membrane mirrors were used in several early head-worn display prototypes. Kramida's 2016 review of the vergence-accommodation conflict describes a typical MOEMS deformable membrane mirror (DMM) as a thin circular membrane of silicon nitride coated with aluminum (or similar materials) suspended over an electrode. The voltage on the electrode changes the curvature of the membrane, which directly refocuses a laser beam scanned onto the retina.

In a varifocal display it is the overall curvature of the mirror, and so its optical power, that is varied. A 2023 review of varifocal MEMS mirrors in Microsystems & Nanoengineering groups the ways of deforming such a thin reflective membrane into pneumatic or hydraulic, electrostatic, thermal and piezoelectric actuation. It notes that the first reported variable-focus mirror (1961) was driven pneumatically by a loudspeaker, and that pneumatic actuation, which controls the pressure in a sealed cavity under the membrane, is still in use. In the 2017 UNC see-through design, the membrane is assumed to deflect uniformly, so that it approximates a spherical concave reflector whose radius sets the depth of the virtual image.

Kramida points out a practical trade-off of these mirrors in head-mounted displays. Tunable lenses, birefringent lenses and sliding optics can be used in a telecentric layout, but a deformable mirror works in reflection, so it needs a more complex off-axis, pupil-forming optical assembly with aberration correction before the light reaches the eye. In return, its optical power can be changed quickly enough to time-multiplex focal depths at 1-100 kHz rates.

History

Astronomy and the origin of adaptive optics

Babcock, a Mount Wilson and Palomar astronomer, suggested in 1953 that if the deviation of rays from all parts of a telescope mirror could be continually measured and fed back to "correct locally the figure of the mirror", both atmospheric seeing and imperfections of the optical figure could be compensated. He judged it impracticable to correct the main mirror itself and proposed instead to place a small ray-controlling element at an image plane conjugate to it. His candidate was the Eidophor, a television projection device in which a thin layer of oil covering a mirror is distorted by electrostatic charge deposited by an electron beam.

According to Davies and Kasper, it took more than 30 years for the technology to support a practical version of Babcock's concept. The US military began investing in adaptive optics in the 1970s and commissioned the first practical system, the Compensated Imaging System, on the 1.6 m telescope on Haleakala, Maui, in 1982. The first astronomical AO instrument, COME-ON, was tested in the late 1980s at the 1.52 m telescope of the Observatoire de Haute-Provence and later installed on ESO's 3.6 m telescope at La Silla. In his review of retinal imaging, David R. Williams credits J. W. Hardy and colleagues with the first empirical demonstration of adaptive optics in astronomy, published in 1977.

Vision science

In 1989 Dreher, working in Josef Bille's laboratory at the University of Heidelberg, used a deformable mirror in a scanning laser ophthalmoscope, correcting only the astigmatism in one subject's eye from a spectacle prescription. At the University of Rochester, Junzhong Liang, Williams and Donald T. Miller then built the first closed-loop adaptive optics system able to correct higher-order aberrations of the eye. It combined a Shack-Hartmann wavefront sensor, a 37-actuator deformable mirror and a flash-illuminated fundus camera. Each measure-and-correct loop took about 15 minutes, and four or five loops were needed. The work was published in the Journal of the Optical Society of America A in November 1997; the authors reported that correcting the eye's aberrations gave normal eyes "supernormal optical quality" and allowed retinal images that resolved structures the size of single cells. Williams later recalled that the only deformable mirror available when he first considered building an adaptive optics ophthalmoscope carried a US$1 million price tag, and that the demonstration had to wait until the price fell by more than an order of magnitude.

Varifocal mirrors in early 3D displays

Flexible mirrors were used for three-dimensional displays long before head-mounted displays. In a 1967 paper in Applied Optics, Alan C. Traub of the MITRE Corporation described a volumetric display in which a vibrating membrane mirror, driven electrostatically or by a loudspeaker, made the virtual image of a two-dimensional pattern sweep through a volume of space; his demonstrations included a simulated air traffic control display. Eric G. Rawson wrote about the technique in IEEE Spectrum in 1969 in an article titled "Vibrating varifocal mirrors for 3-D imaging". The 2023 MEMS review counts these volumetric stereoscopic displays among the first applications of varifocal mirrors and notes that much smaller versions of the same idea have returned in near-eye AR displays.

The "deformable mirror device" at Texas Instruments

The digital micromirror device behind DLP projection started as a deformable mirror. According to the ASME landmark brochure, a Department of Defense-funded project at Texas Instruments, led by Larry Hornbeck from 1977, set out to build a device to modulate light. Its first design used a metalized polymer membrane controlled electrostatically pixel by pixel. In late 1980 Hornbeck filed the first patent for the "deformable mirror device or DMD", in which an analog voltage across an air gap deformed a metalized membrane. By 1981 he had moved away from membranes toward all-metal reflective cantilever micromirrors integrated with an address circuit on a silicon substrate. In 1987 he began using micromirrors as on-off switches, the basis of the digital micromirror device. A present-day DMD instead uses millions of hinged microscopic mirrors to direct light through a projection lens.

Applications in VR and AR

Conventional stereoscopic headsets show every pixel at one fixed optical distance, so the eyes converge on a virtual object's depth while focusing at a different one. Deformable mirrors are one of several tunable elements studied for removing that mismatch; others include focus-tunable lenses, liquid crystal lenses, birefringent lenses and sliding optics.

Retinal scanning displays (University of Washington)

At the Human Interface Technology Laboratory (HITLab) of the University of Washington, Sarah C. McQuaide, Eric J. Seibel, John P. Kelly, Brian T. Schowengerdt and Thomas A. Furness placed a MEMS deformable membrane mirror in the optical path of a monocular retinal scanning display, a form of virtual retinal display. Kramida reports that the system produced a continuous focus range from 33 cm to infinity (3 to 0 diopters). Schowengerdt and colleagues then made the design binocular by splitting the laser beam into left and right images, extended the focal range to 0-16 diopters, beyond the accommodation range of the human eye, and added beamsplitters at the exit pupils to show that such a display could be used for AR. Kramida describes these as bench proof-of-concept systems that displayed simple line-pair images; autorefractor measurements showed that viewers' accommodation followed the intended focal depth. Schowengerdt and Seibel also synchronized the membrane's curvature with content at two depths, frame by frame, to produce a frame-sequential multiplanar image, before turning to arrays of scanning fiber projectors. In 2016 GeekWire described Schowengerdt as a longtime University of Washington professor and the chief science officer and co-founder of Magic Leap.

Multi-focal-plane optical see-through display (University of Arizona)

Xinda Hu and Hong Hua built a monocular optical see-through head-mounted display whose image generation unit combined a DMD display, a deformable membrane mirror, additional lenses and a polarization beam splitter. The mirror switched optical power at up to 1 kHz, synchronized with the display's content for six depths, to produce six evenly spaced focal planes between 0 and 3 diopters. A freeform prism eyepiece with a compensating prism gave a 50 by 45 degree see-through view, with a central 40 degree low-distortion region for virtual imagery. The 2014 Optics Express paper reports an angular resolution of 1.8 arcminutes across a 40-degree diagonal field of view in the virtual display path and 0.5 arcminutes for the see-through view.

See-through deformable membrane mirrors (UNC, NVIDIA and MPI)

In 2017 David Dunn, Henry Fuchs and colleagues from the University of North Carolina at Chapel Hill, the Max Planck Institute for Informatics, Saarland University and NVIDIA Research made the deformable mirror itself the see-through optical combiner. Each eye has one varifocal membrane mirror that reflects a display mounted above the eye, and an eye tracker sets the target depth. The authors give the benefits as a 100-degree diagonal field of view and depth switching from 20 cm to optical infinity within 300 ms; the measured monocular field of view of the prototype was 60 degrees horizontal and the binocular field 90 by 45 degrees. The paper received the best paper award at IEEE VR 2017.

The membranes were made in-house: polydimethylsiloxane (PDMS, Dow Corning Sylgard 184) was spin cast to a thickness of about 240 micrometers, cured, coated with a 20 nm silver film and stretched across a 3D-printed vacuum housing. Each eye's housing was held at a controlled partial vacuum by a regulator able to set pressures between -1.3 and -80 kPa, which gave about 60 addressable stable depth planes from 0.2 to 7 diopters. An infrared LED and camera above each mirror measured its shape for closed-loop control, and an extra lens reduced the astigmatism caused by the elliptical aperture. Road to VR described the work as very preliminary. The authors list its limitations: it shows only one depth at a time, the membrane's change of shape takes up to 300 ms in the most extreme case, and the optical design is bulky.

A 2018 follow-up by Dunn, Praneeth Chakravarthula, Qian Dong and Fuchs, which took first place in the SPIE Digital Optics for Immersive Displays student optical design challenge, used the membrane as a deformable beamsplitter with no additional optical elements. It reported a monocular field of view of 75.5 by 70.6 degrees, focal distances from infinity to less than 10 cm, 4-6 cycles per degree of angular resolution, and a head-mounted unit of 5.5 by 12.5 by 15.2 cm weighing 452 g including unoptimized mounting hardware. Instead of an air compressor and pressure regulators, an 8-inch subwoofer modulated the air pressure in each eye's membrane housing, with a pressure sensor providing feedback.

Summary of prototypes

Year	Group	Mirror and role	Reported focus range and speed
2003	University of Washington HITLab (McQuaide et al.)	MEMS deformable membrane mirror in a monocular retinal scanning display	Continuous, 3 to 0 diopters (33 cm to infinity)
2003	University of Washington HITLab (Schowengerdt et al.)	MEMS deformable membrane mirror in a binocular retinal scanning display with see-through beamsplitters	0 to 16 diopters
2014	University of Arizona (Hu and Hua)	Deformable membrane mirror in the image generation unit of a freeform optical see-through display	Six focal planes from 0 to 3 diopters, switching at up to 1 kHz
2017	UNC, MPI Informatik, Saarland University, NVIDIA (Dunn et al.)	Pneumatic PDMS membrane mirror as see-through combiner, gaze-driven	20 cm to infinity within 300 ms
2018	UNC (Dunn et al.)	Pneumatic deformable beamsplitter driven by a subwoofer	Infinity to less than 10 cm

Open problems

The 2023 MEMS review notes that deformable MEMS mirrors have reached an accommodation range of 0 to 14 diopters in retinal scanning displays. It also says that see-through deformable beamsplitters, though similar in concept to Rawson's 1969 work, ask more of varifocal MEMS mirrors: transparent substrates, larger diameters and large optical powers. Kramida's review adds the off-axis optics a reflective element requires as a design cost compared with lens-based varifocal approaches.

See also

• Varifocal display
• Vergence-accommodation conflict
• Near-eye light field display
• Depth cue
• Accommodation (eye)
• Microelectromechanical systems
• Spatial light modulator
• Optical aberration
• Liquid crystal lens
• Volumetric display