A display module can leave the panel fab with a clean inspection report and still ship inside a defective pair of glasses. Magnified through near-eye optics, a brightness variation that would never survive a monitor review becomes a fixed blotch hanging in a 40-degree virtual screen, and an efficiency-charged optical stack makes it worse: documented module measurements put Fresnel transmission near 80 percent, pancake, freeform, and birdbath modules between 15 and 35 percent, and diffractive waveguides near 0.3 percent — so the panel runs harder, closer to its drive limits, where non-uniformity and color shift grow. Defect acceptance, not defect absence, is what a sourcing team actually controls. This guide is the fifth piece in our display-and-optics series, after the birdbath, waveguide, pancake, and fast-LCD entries, and it covers the four defect families that decide whether a display lot ships: mura, image retention, ghosting, and flicker.
Key Takeaways
- Magnification, not panel grade, is what makes near-eye defects visible: a 0.49–0.70-inch micro-OLED magnified across a wide field shows sub-pixel luminance error a phone panel never would, and off-axis emission adds peripheral color shift as field of view widens.
- Uniformity needs multi-gray testing, not one full-white number: documented near-eye module tests pair 9-point luminance uniformity with checkerboard contrast (a measured module reports 34:1) and MTF falloff from about 66 percent at 2.67 lp/degree to under 17 percent at 13.35 lp/degree.
- Retention is specified by recovery, not by denial: OLED luminance decays with cumulative use, so acceptance means exposing a static pattern at defined luminance for a defined duration, then requiring the residual to clear a percent threshold within a stated recovery window.
- Ghosting is response time against the frame budget: 90 and 120 Hz leave 11.11 and 8.33 ms per frame, overdrive fast-LCD clears grey-to-grey under 1 ms, and micro-OLED responds in microseconds — the acceptance variable is the GtG distribution at temperature, not a best-case datasheet number.
- Flicker is designed out, then verified: perception research places the smoothness threshold near 90 Hz and perceptual saturation around 200 Hz, so dimming frequency belongs in the specification, with a camera-banding check when the glasses will appear on camera.
Why Near-Eye Magnification Rewrites Defect Standards
The panels behind current glasses are small and extremely dense. Viewing glasses run 0.49–0.70-inch micro-OLED panels — the 0.68-inch class documented at 1,920 × 1,080 and 108 percent sRGB — while VR-class headsets use 1.03-inch panels at 2,560 × 1,440 per eye, 3,514 PPI, and 120 Hz, and flagship spatialcomputing headsets push past 4,000 PPI at roughly 4K per eye. None of those numbers is the point. The point is that near-eye optics magnify a stamp-sized active area until it fills the visual field, so defects that sit below the visibility threshold on a six-inch phone panel — a dark sub-pixel, a five-percent luminance dip, a slow row — become stationary marks that track with every head movement. The screen-door effect is the pixel-level version of this; mura, retention, ghosting, and flicker are its luminance, aging, motion, and temporal cousins.
The optical stack compounds the problem. With Fresnel transmission near 80 percent, a panel needs roughly 188 nits of output for 150 nits at the eye; birdbath and pancake modules at 15–35 percent efficiency need panel outputs in the 750–2,667-nit range for 150–400 nits in-eye; diffractive waveguides near 0.3 percent efficiency demand panel luminance several orders of magnitude higher. Driving emitters harder steepens their luminance response and widens pixel-to-pixel variation, and micro-OLED emission angle varies with viewing position — documented as peripheral color shift and edge luminance roll-off as field of view widens. A panel specification that never mentions angular behavior is describing a different product than the one wearing optics.
Mura: What a Single Uniformity Number Hides
Mura is spatial luminance or color non-uniformity visible on flat fields, typically worst at low gray levels. In micro-OLED it accumulates from three physical sources: pixel-to-pixel emitter variation on the organic layer, threshold variation in the silicon backplane TFTs that set per-subpixel current, and cumulative aging that diverges across the panel once fixed-content patterns start wearing some regions harder than others. Fast-LCD has its own versions — backlight spreading and cell-gap variation — which is one reason the fast-LCD sourcing guide treats mura classes as an acceptance item rather than a datasheet line.
The measurement discipline is what separates a real acceptance limit from a checkbox. Documented near-eye module qualification pairs a 9-point luminance uniformity sweep across the field with system-level metrics: checkerboard contrast — one measured module reports 34:1 — and MTF measured in line pairs per degree, falling from 66 percent contrast at 2.67 lp/degree to the 14–17 percent band at 13.35 lp/degree. Uniformity measured only at full white hides exactly the low-gray blotches that magnification exposes; angular color uniformity measured only on-axis hides the edge shift that wide-field optics create. Unit-level compensation data — the de-mura map applied per module — is baseline practice in micro-display production; what varies between programs is whether the before-and-after maps, the gray sweep, and the thermal-soak retest are archived and enforceable when a lot arrives out of spec.
Practical spec language therefore has four parts: the gray fields tested (including low gray), the measurement zones (center and periphery), the limit in percent per field and zone, and the evidence attached to every rejected and accepted lot. Limits written as adjectives — uniform, clean, no visible mura — are not limits.
Image Retention and Burn-In: Specify Recovery, Not Hope
OLED luminance decays with cumulative use, and it decays unevenly when the content is uneven. Video glasses that spend their life letterboxed at 16:9 or 2.39:1 age the picture area harder than the bars; glasses with a persistent status row or a fixed crosshair age that strip fastest. The visible result is retention — a faint ghost of the static content — and over a long enough exposure, a permanent luminance delta. This is a physics statement about organic emitters, not a quality statement about any fab, and it is why retention belongs in the same acceptance framework as uniformity rather than in a footnote.
An honest retention specification is built from a burn-and-recover cycle: display a defined static pattern at defined luminance for a fixed duration, then measure residual luminance difference at set gray fields after a stated recovery window. The pass condition is a percent threshold attached to the recovery time, because retention that clears predictably within seconds behaves differently in the field from retention that lingers for the rest of the session. Firmware mitigations — pixel shifting, idle dimming, auto-brightness, and content-adaptive compensation — change the aging curve and belong in the same quality plan as the measurement. The industry-standard lifetime figures used at panel level (T50-class decay metrics) are the fab’s own qualification inputs; what your spec needs is the module-level recovery cycle, run on your content profile, before the design is locked.
Ghosting: Response Time Against the Frame Budget
Ghosting is what happens when pixel response cannot keep up with the frame cadence. At 60 Hz a frame lasts 16.67 ms, at 90 Hz 11.11 ms, at 120 Hz 8.33 ms — and motion-to-photon latency needs to stay inside roughly 20 ms for comfortable use. A liquid-crystal pixel still mid-transition when the next frame arrives smears the old image into the new one. The response ladder across technologies is well documented: standard LCD sits at multiple milliseconds, overdrive fast-LCD brings grey-to-grey under 1 ms, micro-OLED responds in microseconds, and micro-LED in nanoseconds. Panel class buys you the ceiling; the acceptance question is what the distribution looks like at temperature.
That is the near-eye twist. Low persistence — pulsing the backlight or inserting black frames — is how fast-LCD headsets buy motion clarity inside a fixed frame budget, and it trades against brightness: one documented mainstream headset runs 100 nits in its low-latency mode. Acceptance data therefore needs the GtG transition matrix across the gray range, retested at hot and cold soak, alongside the persistence setting actually shipping in firmware. A single best-case GtG number on a datasheet is a marketing sentence; the distribution is an engineering one.
Flicker: Design It Out, Then Verify With a Camera
Flicker perception has two useful anchors from vision research: around 90 Hz is where most observers stop seeing discrete refresh as stutter, and around 200 Hz — a 5 ms refresh — is where effective perceptual saturation sits for near-eye use. Dimming is where flicker re-enters: PWM dimming at low duty cycles can push visible modulation below both thresholds, which is why the dimming frequency belongs in the display specification next to the refresh rate, not inside a driver IC datasheet nobody on the buying team reads. DC dimming sidesteps the issue at some cost in low-brightness control precision, and the choice between them is a design decision that should be made explicitly per product.
One verification is routinely forgotten: camera banding. Glasses that appear in front of cameras — streaming, video calls, drone footage — can look flicker-free to the eye while a rolling-shutter sensor records scrolling bands, because the sensor samples the frame differently than your retina does. Shipping refresh rates make this concrete: current micro-OLED products run 60–120 Hz, with 90–120 Hz typical and silicon backplanes documented above 140 Hz, so the banding check at the shipping dimming duty cycle is a five-minute test that prevents a launched product from flickering in every video review of it.
The Acceptance Table Worth Sending With an RFQ
Every quantity below is routinely measured in module qualification. The difference between programs is not whether suppliers can measure it — it is whether the limits are numeric, conditioned, and attached to lot-level evidence your team can audit.
| Defect family | Primary metric | Unit | Verification method | Spec hook |
|---|---|---|---|---|
| Mura | Luminance uniformity at defined gray fields, by zone | % | 9-point or camera-based gray sweep, retest after thermal soak | Lot acceptance, per module |
| Retention | Residual luminance delta after static exposure | % at stated recovery time | Burn-and-recover cycle on program content profile | Life-test and firmware plan |
| Ghosting | GtG distribution across gray transitions at temperature | ms | Transition matrix at hot and cold soak, shipping persistence setting | Panel and driver acceptance |
| Flicker | Dimming frequency and banding onset at shipping duty cycle | Hz | Photodiode trace plus rolling-shutter camera check | Design and firmware acceptance |
Write each limit as a number with its conditions attached — gray field, zone, temperature, recovery window, duty cycle — and the table becomes the contract. Write the same limits as adjectives and every dispute becomes an opinion.
How VISGLASS Handles Display Acceptance
Display modules in our VR and video-glasses programs are assembled in a class-100 optical workshop, and acceptance evidence is treated as product data: uniformity maps at multiple gray levels, retention burn-and-recover cycles, GtG distributions at temperature, and photodiode flicker traces are maintained per module and written into the project quality plan, so the limits your team sets are the limits the lots meet — and the evidence is auditable when you need it. For mid-size programs the same measurement set is available at pilot volumes, which is usually where the expensive acceptance mistakes are caught.
Running a display acceptance review for a glasses program? Share the defect limits you are considering — or ask which limits make sense for your content profile — and we will give you a straightforward read on what is achievable at your volumes.
FAQ
Q1: What is mura in VR and video glasses?
Mura is spatial luminance or color non-uniformity visible on flat fields, usually worst at low gray levels. In micro-OLED it comes from pixel-to-pixel emitter variation, silicon-backplane TFT variation, and uneven cumulative aging; magnified by near-eye optics it reads as blotches or streaks hanging in the image. It is managed with per-module compensation and verified with multi-gray, multi-zone uniformity measurement — a full-white check alone does not see it.
Q2: How is image retention tested before mass production?
A defined static pattern is displayed at defined luminance for a fixed duration, then residual luminance difference is measured at set gray fields after a stated recovery window. The pass condition is a percent threshold attached to the recovery time, run on content that resembles the shipping use — letterboxed video, persistent UI rows, fixed crosshairs. Firmware mitigations such as pixel shift, idle dimming, and content-adaptive compensation are validated in the same cycle.
Q3: Can VISGLASS provide defect acceptance data for custom glasses programs?
Yes. Uniformity maps across gray fields, retention cycles, response-time distributions at temperature, and flicker traces are part of the per-module records from our class-100 optical assembly, and the limits can be written into the project quality plan so lot-level evidence is auditable. Share your defect limits or your content profile and we will map what is feasible at your volumes.