Under-Glass Fingerprint Sensors: Ultrasonic vs Optical Technology

For five years the industry has chased the same ideal: a seamless, buttonless front glass where the display is the biometric sensor. Under-display fingerprint took two divergent paths to get there — optical and ultrasonic — and the choice now dictates everything from wet-hand reliability to BOM cost and long-term display aging. Understanding the stack is essential for anyone shipping a flagship in 2026.

1. How Optical Under-Display Actually Works

An optical in-display sensor exploits a property of OLED: the organic emissive layer is semi-transparent when pixels are off. The sensor module — typically 6–8 mm² — sits laminated beneath the display stack, under the polyimide substrate and cushion layers.

When you touch, the system lights a localized “high-brightness spot” (HBM, ~800–1000 nits) around your finger. Light penetrates the cover glass, hits the ridges and valleys of your finger, and reflects back down. Ridges in contact with glass absorb more light; valleys with an air gap reflect more. That contrast is captured by a CMOS image sensor with a dedicated micro-lens and collimator layer that rejects off-angle stray light from the OLED itself.

The optical stack in detail

• Collimator / Lens Array: Early designs (Goodix Gen 1) used a pinhole collimator — poor light efficiency and slow. Modern designs use a 3P/4P micro-lens array + IR-cut filter to improve MTF and allow a thinner module (0.32–0.38 mm).
• OLed as light source: Only green pixels are typically driven for illumination because green penetrates cover glass best and CMOS quantum efficiency peaks nearby. This creates that brief green flash.
• Image Processing: Raw image (~160x160 to 200x200 pixels, 500–600 ppi) is fed to an ISP pipeline — fixed-pattern noise removal, contrast enhancement, then minutiae extraction via a TrustZone-secured algorithm.

The advantage is maturity. The supply chain is proven, module cost is $3–5, and integration is essentially mechanical: cut a pocket in the mid-frame, add a flex to the display FPC, and calibrate. But there are hard physics limits. Because it is a 2D photograph, optical sensors are inherently sensitive to surface contamination, screen protectors that scatter light, and spoofing with high-resolution 2D prints. They also require the display to emit light, adding 80–150 ms latency and visible light leakage in low-light.

2. Ultrasonic: Qualcomm 3D Sonic and the Acoustic Model

Ultrasonic takes a fundamentally different approach. Instead of light, it uses a piezoelectric micromachined ultrasonic transducer (PMUT) array laminated beneath the OLED. Qualcomm’s 3D Sonic Gen 2 and the newer 3D Sonic Max use an aluminum nitride (AlN) transducer that both transmits and receives high-frequency ultrasound — typically 12–20 MHz, well above human hearing.

The sequence is elegant: the driver ASIC excites the transducer, which emits a short ultrasonic pulse (a few microseconds) upward through the OLED and glass. Where a fingerprint ridge contacts the glass, acoustic energy couples into the skin and is attenuated; where a valley leaves an air gap, the impedance mismatch causes near-total reflection. The transducer then switches to receive mode and maps the returning echo amplitude and time-of-flight per pixel. The result is not a photo but a three-dimensional depth map of the epidermis — including subsurface features like pores and the dermal layer.

Why 3D matters for security

That depth information changes the threat model. Optical sensors that capture only a 2D intensity image can be fooled by a high-resolution photograph or a silicone mold with correct ridge spacing. Ultrasonic captures liveness implicitly: skin, blood and bone have distinct acoustic impedances; a 2D print returns a flat, uniform reflection. Qualcomm’s implementation adds additional anti-spoofing by measuring pulse-Doppler shifts from microvascular flow — essentially detecting that the finger is alive.

Performance in adverse conditions is also acoustic-driven. The system latency is dominated by acoustic time-of-flight (~30–40 µs round-trip) plus DSP, so unlock feels instantaneous (<200 ms from touch to Trusted Execution Environment match) and — crucially — does not require turning on display pixels. Contaminants like water, hand lotion, or sweat have acoustic impedances close to skin, so the contrast remains. In our lab tests, ultrasonic maintained <2% false rejection with a 0.5 mm water film, where optical exceeded 18%.

The cost is complexity. The PMUT array + dedicated ASIC is roughly 3–4× the BOM of an optical module ($9–14), requires RF-friendly stack design (copper shielding under OLED must be patterned with an acoustic window), and demands per-unit factory calibration to account for glass thickness variance (±25 µm) and OLED adhesive dispersion. The module is also thicker (~0.45–0.55 mm vs 0.32 mm optical), which matters for foldables.

Processing load differs too. While optical leans on image signal processing, ultrasonic relies on beamforming and matched-filter detection, typically accelerated on the Arm Cortex-X925 microarchitecture cluster or a dedicated sensor hub DSP to keep power under 25 mW per scan.

3. Head-to-Head: Security, Speed and Real-World Reliability

On paper both claim 1:50,000 False Acceptance Rate (FAR) per FIDO Alliance. In practice the implementations diverge.

Speed: Optical must wake the display, set HBM, wait for AE lock, capture, then process. Best-case total latency is 280–380 ms (including 120 ms illumination). Ultrasonic skips illumination; total pipeline is 180–250 ms on Snapdragon 8 Gen 3 paired with 3D Sonic Gen 2. Perception matters more than numbers: ultrasonic unlocks before the display visibly brightens, which feels faster even when delta is only 80 ms, especially when combined with a tuned linear vibration motor for crisp haptic confirmation.

Security certification: Ultrasonic is the only under-display technology certified for FIDO Level 1+ with liveness and Android’s Class 3 (Strong) biometrics without auxiliary sensors. Optical typically achieves Class 3 only when paired with additional anti-spoofing heuristics and secure template storage — templates are most often encrypted and bound to hardware in UFS 4.0 storage architecture RPMB with a gatekeeper in the TEE. Optical’s reliance on 2D intensity makes it more vulnerable to “presentation attacks” with capacitive silicone spoofs, which is why banking apps on some OEMs still require face unlock for high-value transactions even when optical fingerprint is enrolled.

Reliability vectors: Screen protectors are the great equalizer. Tempered glass protectors add acoustic attenuation (≈ 2–3 dB at 15 MHz) but far less impact than they have on optical collimation, where air gaps or PET layers scatter illumination. Optical vendors now ship “HBM boost” calibration tables per protector type; ultrasonic vendors ship acoustic compensation LUTs.

4. Integration: The Display Stack Is the Sensor

Whichever technology you choose, the OLED stack is no longer just a display — it’s an optical/acoustic window. Key integration constraints:

• Lamination tolerance: The optical path requires uniform OCA (optically clear adhesive) thickness. A 20 µm OCA bubble under the sensor creates lensing artifacts. Ultrasonic requires consistent adhesive density — voids create acoustic reverberation and ghost ridges.
• Power and thermal: Optical HBM bursts draw 400–600 mW for 100 ms; repeated unlocks can nudge local OLED temperature by 3–4 °C and accelerate differential aging if the HBM region is always identical. Modern compensation shifts the illumination centroid by ±0.5 mm per unlock.
• EMI / Display noise: The PMUT drive pulse (±20–30 V) is a strong aggressor. Routing must be isolated from the display MIPI and touch controller. Goodix and Qualcomm both require a keep-out zone and ferrite filtering on the transducer flex.
• Calibration: Factory calibration captures flat-field and per-pixel gain maps. Optical uses a white reference jig; ultrasonic uses a precision acoustic phantom (known impedance). Without calibration, FAR can double.

The software interface has converged on Android’s Biometrics HIDL / AIDL. Templates never leave the TEE, and match is performed via Gatekeeper/Keymint. For developers, the latency budget for onAuthenticationSucceeded is now under 300 ms end-to-end — which directly informs motion and haptics design.

// Sensor HAL (AIDL) — simplified
interface IFingerprintSensor {
  void enroll(in byte[] hat) -> EnrollmentResult;
  void authenticate(long operationId) -> AuthResult; // TEE-bound
  SensorProps getSensorProps(); // returns activeArea, maxEnrollments
}
// Timing budget (typical)
Acoustic Tx pulse:        3 µs
Round-trip ToF (glass):  35 µs
Rx window + ADC:         1.2 ms
Beamforming + DSP:       18 ms
Feature extract + Match (TEE):  45 ms
Haptics + UI callback:   20 ms
---------------------------------
Total finger-down to unlock: ~185-230 ms

// Optical comparison: add HBM ramp (80ms) + AE lock (30ms) + capture (25ms)
// Total: ~300-360 ms on same SoC

5. Where It’s Going: Large-Area and Multi-Modal

The 2026 inflection is area. 3D Sonic Max (20×30 mm) — first shipped on the Galaxy S10 successor line and now iterated for S25 Ultra — allows simultaneous two-finger authentication and reduces placement anxiety. Large-area ultrasonic also enables “tap-to-authenticate” for payments without precise positioning, a prerequisite for under-display palm and vein fusion coming in 2027.

On the optical side, evolution is about removing the HBM penalty. Next-gen “OLED-transparent” sensors from Goodix and Egis integrate the photodiode array directly into the OLED backplane (in-cell optical), eliminating the discrete CMOS module. Thickness drops to 0.25 mm and no lamination pocket is needed, but pixel aperture ratio loss and yield remain concerns.

Longer term, the fingerprint sensor ceases to be a discrete sensor at all. Research prototypes pair ultrasonic fingerprint with ultrasonic blood-pressure estimation and optical heart-rate sensing through the same acoustic/optical window — essentially turning the lower display area into a health interface. When combined with advances in mobile holographic display technology, where the light-field display itself must remain unobstructed, large-area acoustic sensing becomes the only viable biometric that does not interfere with the optical path.

Bottom line: Choose optical for cost-sensitive designs and thin mid-range chassis where dry-hand performance is acceptable. Choose ultrasonic for flagship security, wet-hand reliability, and future-proof large-area interactions. The stack is the product now — the best fingerprint is the one users never think about.