Introduction

In September 2026, I gave a “Master Class” at MicroLED and ARVR Connect on the conference’s first day. I tried to cram about 2 hours of content into a 1-hour presentation. Using this presentation as a starting point, I have added considerably more information and figures to this multipart series.

In Part 1, I covered what I see as the major AR Challenges facing companies in 2026. In Part 2, Waveguide Trends, I discussed the trends I noticed in 2026 for both Reflective and Diffractive waveguides.

In this part, I discuss metastructure combiners from TDK and Fourier Optics, along with some of TDK’s laser beam scanning (LBS) activity. I also briefly mention CREAL’s HOE combiner. I put some background information on “Free Air Combiners” in the Appendix.

I also want to mention more “radical” combiner approaches, including Gixel’s Active Mini-Mirrors and Smith Display’s non-linear crystal combiner.

As so often happens in these articles, this started as a short piece on TDK’s Meta-Optics Mirror and Fourier Optics’ Metasurface, but it kept growing as I dug deeper and added related technologies and background information.

Metasurface Definition

I also want to note that “Meta Optics” and “Metasurfaces” are types of flat optics made with subwavelength nanostructures and should not be confused with the company Meta. Often, the diffraction gratings used in waveguides have similar minimum dimensions but are typically linear in one dimension, whereas metasurfaces may have a more complex two-dimensional structure. By one definition:

A diffraction grating is a traditional optical component with a periodic groove or slit structure that disperses light into different wavelengths based on classical diffraction and interference principles. A metasurface (or optical metastructure) is an engineered, subwavelength-scale flat array of tiny nano-antennas that manipulates light phase, polarization, and amplitude at a scale smaller than the wavelength of light itself.

Many diffraction gratings aren’t simple linear gratings with uniform spacing; they often have substructures that change the grating’s optical characteristics. For example, see the Magic Leap patent application (right), which describes complex structures with feature sizes down to 100nm, or about 1/6 the wavelength of green light. So, there seems to me to be a continuous spectrum between what is called a diffraction grating and a metasurface.

AI XR Podcast, Display Skeptics, CES 2027 and SPIE AR/VR/MR 2027

Last week, I was interviewed on the AI/XR Podcast (on YouTube and as a podcast) with hosts Charlie Fink (Disney legend, Forbes), Ted Schilowitz (Red Camera), and Rony Abovitz (Magic Leap founder). The interview covers my background and my views on today’s AR landscape.

Radu Reit and I publish monthly AR Updates, along with longer videos on various topics, on our Display Skeptics YouTube channel. We just recorded our November update, and it should be on YouTube in a few days (as of this writing).

Looking ahead, I’m planning to attend CES in early January 2027 and SPIE’s AR/VR/MR conference in early February 2027. If you want to meet with me at either event, email meet@kgontech.com.

TDK Major Moves in LBS Displays for AR

On October 2, 2026, TDK announced its Meta-Optics Mirror (left), a new approach to a “free-air combiner.” Not long before, in May 2026, TDK acquired Oqmented’s core assets, IP, and people. Oqmented developed a mirror that is sealed in a vacuum for Lissajous Laser Beam Scanning (for more on Oqmented and Lissajous LBS, see: AWE 2021 Part 2: Laser Scanning – Oqmented, Dispelix, and ST Micro). TDK has had a long-standing relationship (since at least 2020) with laser maker QDLaser, and in June 2026, they entered into a business cooperation agreement to develop retinal-projection optical engines for smart glasses.

I was aware of QD Laser, which spun out from Fujitsu, back when I worked on laser Illuminate LCOS in 2010 and 2011 (see my SID 2011 paper on the subject), and I have seen QD Laser’s LBS direct retinal scanning several times, at least as far back as 2016 (right).

Before the acquisition, Oqmented seemed focused on waveguide-based LBS, while TDK’s new Meta-Optics mirror supports direct retinal projection LBS. When LBS uses waveguides, the resulting image light from the waveguides’ pupil expansion is no longer “Maxwellian” or focus-free and, from a projection perspective, behaves more like a flat-panel display.

This blog has discussed LBS direct retinal projection (LBS-DRP) many times, including those by Microvision, Bosch, QD-Laser, North Focals, Intel (Vaunt), and Eyeway Vision. I have also covered LBS use in waveguides with the HoloLens 2 and Oqmented.

The waveguide’s pupil expansion lets you see the image over a much wider range and eliminates issues with a free-air laser path between a temple projector and the eye (see appendix for more on free-air combiners). Anything from dust on a surface to a hair crossing the free-air path can cast a large, sharp shadow that blocks part of the image. Free-air laser projection also raises perceived eye-safety concerns. At least for now, the TDK website shows them continuing with Oqmented’s waveguide approach (see below).

So TDK appears to be pursuing both free-air and waveguide-based LBS AR glasses development.

TDK’s Meta-Optic Mirror

TDK’s Meta-Optic Mirror uses a nano-scale structure (ranging from 200nm to 1 micron) that redirects light from an LBS scanner toward the eye. The image from TDK (right) shows why the special structure is necessary when projecting light from the temple of glasses. The light from a simple mirror would bounce at an angle that would miss the eye (for more, see Free-Air Combiners with Temple Projector in the appendix).

The nanostructure is made in a semiconductor-like process on a thin substrate that can be embedded in an “ordinary” eyeglass lens. The Meta-Optic mirror is likely sandwiched between two lens halves, with a mirror cavity, as waveguides are done today. In the figure below left, the laser projector output appears to go nearly parallel to the frame to a second mirror near the lens, then at a very shallow angle to hit the nano-scale reflector, which greatly reduces the free-air path compared to the conceptual figure (above right).

This current meta-optics design is for monochrome (green only). It’s reasonable to assume the nanostructure size and spacing might have some wavelength dependence, as in diffractive waveguides. TDK’s website states, “While the current system supports monochrome images, the company is accelerating preparations to demonstrate full-color operation and achieve mass production within the next few years.”

Oqmented Coherent (Progressive) Laser Scanning (Similar to Meta)

In AWE 2021 Part 2: Laser Scanning – Oqmented, Dispelix, and ST Micro) I discussed some of the problems with LBS in general and Lissajous scanning in particular. At MicroLED and AR/VR/MR Connect 2026, I saw a short working demo of what Qqmented (now TDK) originally called “Progressive Lissajous Scanning” (below left). The TDK person said they developed this concept independently of Meta. But with Meta discussing a very similar concept at SPIE AR/VR/MR 2025 in “Advances in MEMS-based laser scanning displays for AR glasses” (below right), it appears TDK has since adopted Meta’s nomenclature on TDK’s website.

With Lissajous scanning, a single mirror is driven by two sine waves, one for horizontal and one for vertical, creating a figure-8-like sweeping pattern in the laser beam(s). The pattern’s shape depends on the ratio of the horizontal and vertical frequencies. Typically, the horizontal direction is faster than the vertical one; in Oqmented, the horizontal scan can range from 20 kHz to 40 kHz. With “raster scanning” LBS, the vertical scan typically runs at only about 60 to 120 Hz, whereas the horizontal frequency is on the order of 20 to 60KHz or almost 1,000 times faster (see: HoloLens 2 First Impressions: Good Ergonomics, But The LBS Resolution Math Fails!). This high horizontal-to-vertical scanning ratio makes it “raster-like” (the scan lines are still slightly curved). With Lissajous scanning, the slow (often vertical) scan frequency isn’t that much slower than the horizontal, so the scanning follows a much more curved diagonal pattern (see below right).

Traditional Lissajous Issues

With traditional Lissajous scanning, you have to map an image onto a highly distorted scan. Furthermore, the speed of the beam sweeping varies between zero at the outsides and its fastest speed in the center; the laser has to be modulated to compensate for the change in velocity (see: AWE 2021 Part 2: Laser Scanning – Oqmented, Dispelix, and ST Micro for a more detailed explanation). In essence, the process determines when and where the laser beam will cross the given pixel location, then drives the laser based on both the pixel’s brightness and the beam’s velocity at that time. Thus, pixels are written in a complex pseudo-random process. The human vision system (roughly put) takes a series of snapshots and stitches them together into what we see as a single image, but with the eye’s saccadic movement and the lack of persistence of the laser-scanned image, the eye will occasionally see only part of the scanned image and see the Lissajous scanning.

Coherent Lissajous

Coherent Lissajous uses the same laser scanning as traditional Lissajous, but instead of drawing pixels pseudo-randomly, it draws them in the order the beam sweeps, somewhat like raster-like laser scanning; only the scan lines are more curved and have a steeper diagonal angle. I marked up the image on the TDK website to illustrate the process (right). I arbitrarily labeled one sweep as scan line “1” below where pixels are displayed. That line enters blanking 1B, where the beam turns off during the very slow, curved portion of the sweep as it changes direction. The beam is turned on (displays pixels) for the partial sweep “2” until it hits the bottom of the image, then turns off at 2B. The process continues, forming these curved raster-like segments.

Coherent Lissajous scanning addresses the problem of saccadic eye movement causing the image to break up, as discussed in AWE 2021 Part 2: Laser Scanning – Oqmented, Dispelix, and ST Micro, but I don’t see how it can solve other issues. Fundamentally, there is a resolution problem: pixels from a rectilinear grid get remapped onto a curved scanning process that never exactly hits a single pixel; instead, it crosses near several pixels (see Apple Vision Pro’s (AVP) Image Quality Issues, particularly the appendix on Nyquist). This problem alone reduces effective resolution by about a factor of two in each direction (basic Nyquist theory). I’m also not convinced the scanning process is fast enough to avoid flicker, especially with higher-resolution images. To avoid flicker for most people, the refresh rate (the worst-case time to return to a given pixel) needs to exceed ~90 Hz; otherwise, many people will perceive flicker. In my experience, every LBS scanning system ends up violating the flicker frequency to regain some resolution.

I can’t see how the underlying Lissajous scanning process, with a maximum scanning frequency of 40 kHz, can support more than about 300 lines of effective resolution (after resampling losses) at an ~90 Hz refresh rate. Going back to my days working with TV resolution, NTSC television’s horizontal scan frequency is ~15 kHz, and that was for 720×480 lines at interlaced 60 Hz (non-interlaced 30 Hz), and they did not have to resample the image onto a curved grid. I have been studying various LBS scanning processes for over 25 years (search “LBS” on this blog) and run the math multiple times, and as they say, “the math doesn’t math.” In particular, you might find my HoloLens 2 analysis and my analysis of Magic Leap’s fiber scanning interesting.

From a simple, low-resolution image of the Oqmented (now TDK) prototype, coherent Lissajous scanning solves the saccadic eye-movement breakup issue, but I still noticed some flicker. I don’t have all their specifications, but I think it is safe to draw some conclusions about best-case resolution, with the fastest axis at only 40 kHz.

What about multiple lasers per color?

One fundamental problem with scanning mirrors is that mechanical resonance limits them, and the electrical signal can only tweak them so much. Their peak speed depends on the mirror’s mass, the torsion-bar (or other) mechanism(s) that hold it, and other factors. One of these factors is air and Oqmented vacuum seal in their mirror. But as discussed above, their mirrors’ peak speed is only about 40 kHz, which won’t produce very high image resolution at a non-flickering refresh rate.

Over the last 25+ years, people have often proposed using multiple lasers per color to address the scanning-speed limitations of both single- and double-mirror scanning. HoloLens 2 even shipped with two lasers per color.

Meta’s 2025 SID AR/VR/MR presentation suggested that using multiple lasers for each color could improve coherent Lissajous scanning resolution at a given mirror speed. Assuming no other issues (and there are other issues), you could, in theory, multiply the vertical resolution by the number of lasers per color you add. HoloLens 2 with two lasers per color tried to do this and, to a degree, demonstrated the facility. Please see my 2020 article, HoloLens 2 Display Evaluation (Part 1: LBS Visual Sausage Being Made), which details the issues with using two sets of lasers.

Even if the multiple laser beams were ideally angled and spaced with incredible precision, the scanning mirror’s changing angle during the scan would cause the lasers to vary in spacing. Thus, some of the resolution gained from each set of lasers is lost back to “uniformity correction.” Meta showed examples of its uniformity correction with a multi-laser system, and while compensation improved it noticeably, it still was far from great (see below from Meta’s 2025 SPIE AR/VR/MR presentation), and users have been spoiled by technology with great images on their TV, Monitors, and smartphones.

My rough guess is that they might gain about 1.5X in effective (measurable) resolution for every doubling of lasers after correction. One then has to consider the cost of getting many precision-aligned lasers, the processing for scan and uniformity correction, and the power needed to drive both the lasers and the processing. While it’s a fun thing to think about in a research lab, I can’t see this happening in a high-volume application for decades, if ever.

Fourier Metasurface

At SPIE AR/VR/MR 2024, Fourier Optics first went public with their metasurface reflector. In 2024, they were showing a “temple projector” that was “compatible with uLED and Laser.” While they mention laser projection, their artwork at the time clearly shows a MicroLED projector. They do have at least one older patent filing (see Fig. 1 from 12,498,572 right) showing laser projectors, so they may have originally envisioned laser projection but later largely switched to MicroLEDs as the display system.

At a high level, it might seem similar in function to TDK’s Meta-Optic Mirror, but its structure and optical behavior differ greatly; you have to dig a bit to see how different they are.

In their 2026 renderings (right), they no longer go “free air” from the projector to the lens; instead, they clearly use MicroLEDs and embed the projector in the lens corner.

Like TDK’s Meta-Optics Mirror, they embed their metasurface inside the lens, as shown in their US 2025/0334811 patent application (see the figures below on the left, combined with the picture in the middle from their website).

But the figure above on the right shows that Fourier Optics’ metasurface is constructed very differently from TDK’s Meta-Optics mirror. Whereas TDK’s appears to be a single planar surface, Fourier first generates a series of planar surfaces using a semiconductor-like process, then stacks them vertically with a substrate between them, and finally slices them diagonally to create the final device (labeled 16 in Fig. 20A above). After cutting with a diamond saw, they chemically and mechanically polish (CMP) the diagonally cut stack to create a smooth surface. This stack-and-diagonal-slice process is similar, at least in concept, to how Lumus and other glass reflective waveguides are made.

Figures 3B and 3A from their patent application show how they would embed the metasurface in a lens and the light path. While Fig. 3B is not to scale, Fig. 3A appears to be from an optical simulation of an actual design and gives some idea of the finished metasurface’s size.

From an optical results perspective, the Fourier metasurface with a MicroLED display claims an 8 mm by 10 mm eyebox, 96% transparency, 5,000 nits per lumen, and a 20-degree FOV, with none of the eye glow and rainbow effects associated with diffractive waveguides.

I have not seen a Fourier prototype with my own eyes. They published an image through the optics about a year ago of a green MicroLED (right). There is some visible scattered light around the image itself. It’s unclear if this is an inherent issue with the technology or just the result of being an early prototype. Optical prototypes often lack the coatings and precision needed for the technology to show its full potential.

While the process sounds complicated and expensive, Fourier claims on its website (I’m assuming someday in high volume) that the metasurface could be made for about $3, embedded in an ophthalmic lens for about $5, and build the uLED projector optics for about $2, for a total added cost of optics of less than $10.

Gixel Micro Dynamic Mirror

Gixel’s Micro Dynamic Mirror aims to create a near-ideal free-air combiner. The concept is simple: a series of small mirrors embedded in an index-matching fluid, with electrostatically controlled, semi-transparent mirrors steering projected light from a display directly into the eye’s pupil based on eye tracking (see below left). The number of mirrors, and thus the size of the mirror array, scales with the FOV.

The theoretical efficiency advantage is enormous, hundreds to thousands of times more efficient than waveguides. They would avoid waveguide-structure losses and, in theory, direct light only to the eye that will reach the pupil, whereas waveguides produce an eye box that covers wherever the eye moves (see simulated picture above right).

Gixel discussed their concept on SPIE’s March 2026 Fireside Chat with Bernard Kress https://spie.org/conferences-and-exhibitions/spie-online/spie-online-on-demand/ar-vr-mr-xr. As of the March 2026 Fireside Chat, they had only an experimental version of the mirror system (right), not a full prototype. I haven’t seen any through-the-optics images or had a chance to see it in person, although I have talked with Gixel several times.

While Gixel’s most obvious issue is making a manufacturable, controllable micro mirror, my bigger concern with the concept is some physics issues and how it might interact with the human vision system. Gaps will form between the tilting mirrors, particularly during tilting, which will cause diffraction issues and perhaps image discontinuity. The diffraction issue is well known with Texas Instruments’ DLP, which uses much smaller gaps with its microscopic mirrors.

But my biggest reservation is accurately steering the image based on tracking the human eye. Most obviously, there will be some delay between eye movement and mirror movement. Eye movement is complex, with both small, jittery movements and larger, more intentional movements. Human vision is thought to take a series of snapshots, with partial but not complete blanking during eye movement. As I often say about foveated display attempts, “you can know where the eye is aimed at a given point in time, but you don’t know what it sees.”

The human visual system will likely confuse the tracking movement with image movement. I have seen multiple companies claim to have developed dynamic foveated displays with eye tracking, most notably Varjo, which claimed in 2017 they had it working in a lab but never produced a demonstration and only ever produced a “fixed foveated” (non-moving center only) device. Even more ambitious is the previously discussed Eyeway Vision, which demonstrated good eye tracking and a fixed foveated display, but never demonstrated a dynamic/moving foveated display. I can believe some lab experiment with dynamic image movement can work sometimes, but the human visual system will still cross it at least occasionally, resulting in some form of temporal artifacts. I’m not sure, but I suspect the human visual system could prove to be an intractable problem for Gixel to solve as well.

Smith Displays Laser (Illuminated Display) Wave Mixing in a Nonlinear Crystal

Smith Display showed a very interesting poster at SPIE AR/VR/MR 2026 and even gave me a lab-bench-type demo in their hotel suite. While it’s currently big and crude, their bench demo proves the concept works, and they have published through the optics pictures.

Almost all green laser pointers today use non-linear crystals that frequency-double a near-infrared (NIR) laser beam with a wavelength of ~1064nm to produce a ~632nm green. The same process can produce red and blue outputs from NIR lasers that are double their respective wavelengths.

Smith Display’s concept splits the light output from a NIR laser, sending part as a reference beam and the other part to illuminate a spatial light modulator such as a DLP or LCOS. According to Smith Displays, “We use relaying of the image to produce a Fourier transform in the AR spectacle lens.” The reference beam and the modulated image wavefronts then enter a transparent nonlinear optical material. Where the reference beam and the modulated image cross, the two wavefronts interact to project the image, and thanks to the nonlinear crystal’s frequency doubling (wavelength halving), it appears at half the NIR laser’s wavelength, thus in a visible color. Smith Display has demonstrated this with Red, Green, and blue, both individually and simultaneously (i.e., full color), using the same nonlinear crystal (see below).

Smith Display spun out of the Optoelectronics Research Center and the Optical Engineering and Quantum Photonics Group at the University of Southampton. Their current demonstration is still a lab prototype; it is low resolution, the image quality isn’t great, and the NIL has to be thick to support a reasonable eyebox size. Like most novel concepts, they still have a lot to work out, but they have proven the basics work.

When I last met with Smith Display founder Peter Smith, we were at MicroLED and AR/VR Connect in Eindhoven, near ASML, the developers of the only commercial extreme UV (EUV) lithography machines; it turns out their risky, novel approach, which requires a laser beam hitting tin droplets multiple times at 50K+ times a second, was the only one that worked (for a very good video on the history of ASML’s EUV, see Veritssium’s video).

Conclusion

As I have discussed many times, including in MicroLED + AR/VR Connect and Co-Packaged Optics (CPO), much of AR has an unhealthy business ecosystem. Conventional technology today can’t meet the performance and cost requirements for small, lightweight, highly transmissive AR glasses with anything more than a low-resolution, green-only display that won’t win over most consumers. To some degree, it becomes a bit tongue-in-cheek: “When all the rational ideas won’t work, you start on the irrational.”

Maybe one of these will be the breakthrough AR needs, as ASML had with EUV semiconductor lithography. But even if one of these solves the combiner problem, many others remain (see: AR in 2026 Year In Review Presentation (Part 1: AR Challenges)).

Appendix: Free-Air Combiners with Temple Projector

Free-air combiners take light from a projector, typically on the temple, and redirect the light toward the eye. A simple mirror reflects light at an angle equal to the angle of incidence, normal to the mirror surface. In a head-worn system with a temple projector, this means the light will reflect at an angle that misses the eye.

A simple way to correct this is to tilt the mirror, a so-called “butterfly combiner,” so the center of the flat or curved combiner reflects light at a steeper angle, as shown in the diagram (left) and CREAL’s 2023 prototype (below left). But the butterfly combiner takes up space, doesn’t combine easily with other optics like prescriptions, and is aesthetically unappealing. CREAL knew the butterfly combiner was only a temporary expedient and went on to develop its own “holographic optical element” (HOE) (below right).

HOEs are holograms that create the optical effect of a curved and/or tilted mirror while being thin and angled more or less normal to the eye. HOEs can be made on plastic that can conform to the inner curvature of other lenses. CREAL’s 2023 and 2026 prototypes demonstrate the dramatic difference in appearance while producing a similar optical effect.

If the projector is small enough and the FOV is relatively small, you can position the projector so the combiner doesn’t have to be so tilted. Typically this means positioning the projector near the nose bridge rather than on the temple. Everysight uses this approach.

A curved, rather than flat, reflective combiner moves the focus distance farther away and magnifies the image. This can reduce the need for other optics in the system.

TDK’s Meta-Optic Mirror appears to have a similar net optical effect to an HOE; it uses a metasurface to redirect light rather than a hologram.

As discussed earlier with TDK’s Meta-Optic Mirror, temple projectors and free-air combiners have many physical drawbacks.

Optical Property of Free-Air Combiners

The tilted mirror, HOE, and TDK’s Meta-Optic mirror all preserve the source’s light wavefront. CREAL generates a light field image with content at various focus distances that a conventional waveguide would destroy. TDK’s retinal scanning also requires a mirror, not a waveguide, to work and preserve the light’s focus-free (Maxwellian) nature. This is why all retinal scanning displays have used some form of free-air combiners with either mirror coatings, HOEs, or, in TDK’s case, their mirror optics.

Next in this series

I may take a break to cover other topics before returning to this series, but I plan to cover trends in display devices, particularly LCOS, MicroLED, and LBS, next. My current plan for my next article is to explain why Meta’s 100/4/100 $100 per eye for (4) 100 million units is, while a catchy statement and being repeated by many companies, fundamentally a bad idea.