Transforming lidar with anti-reflective VCSELs

Progressing anti-reflective VCSELs from laboratory prototypes to high-volume production is facilitating cost-effective long-range sensing for automotive lidar.

BY DONG LIANG, CHENG ZHANG, ZHITONG LUO, SHOUZHU NIU, PENGFEI ZHANG, SONG LIU, HUIJIE LI, RYAN Z. RAO, LI ZHAO, XIAOCHI CHEN, HANXUAN LI AND YIJIE HUO FROM VERTILITE

For well over 100 years cars have been rolling off production lines. Over this time, there have been many substantial changes, including the introduction of more efficient engines, seat belts for all, and breaking assistance. But the most profound changes are happening now – a switch from engines burning petrol and diesel to power provided by a battery, and the introduction of self-driving cars.

The latter revolution in the automative industry draws on compound semiconductor lasers for the light source for lidar. These chips fulfil the demands of design engineers, who prioritise power, efficiency, and scalability.

For long-range lidar, high brightness – that’s a measure of the concentration of optical power within a minimal area and a narrow angular spread – is the critical performance metric. To maximise the detection range, sensors need emitters that combine a high output power with an extremely low beam divergence.

One candidate for the light source is the VCSEL. It’s a class of laser that’s produced in very high volumes, and is already widely deployed in short-range consumer applications, such as facial recognition. But if it is to be used in lidar, it must provide a significantly higher brightness.

Helping to meet that goal is the introduction of multi-junction structures, which enable higher power densities. However, that’s only a partial solution to the problem. Higher powers must be combined with efforts to address the excessive divergence of the traditional VCSEL architecture. That’s a weakness that restricts the VCSEL in long-distance perception.

To meet stringent brightness and divergence requirements, our company, Vertilite, has developed and then ramped production of the anti-reflective VCSEL (AR VCSEL). A key feature of our novel laser is the AR layer, integrated between the active region and the extended cavity of a traditional extended VCSEL structure.

Critical to realising a low divergence is limiting the number of lateral modes to just a few. In a VCSEL, the number of supported lateral modes is governed by the effective refractive index contrast between the core and the cladding – these two regions correspond to areas inside and outside the light-emitting aperture. As the oxide layers in a VCSEL provide the dominant contribution to the lateral index difference, this contrast is nearly proportional to what is referred to as the oxide confinement factor.

Figure 1. Divergence angle dependence on the oxide confinement factor for aperture sizes ranging from about 20 µm to about 50 µm.

When the AR layer is integrated between the active region, where current-confining oxide apertures sit, and the extended ‘light reservoir’ region, there’s a significant enhancement to the standing-wave electric field intensity within the extended region. This physical modification leads to a dramatic reduction in the oxide confinement factor. In turn, there’s a fall in the number of lateral modes that are allowed, and ultimately a smaller divergence angle.

Precise control over the oxide confinement factor, and thus the beam’s divergence, is realised by adjusting the property and design of the anti-reflection layer. With appropriate choices, we design VCSELs with narrow angles necessary for long-range detection.

To demonstrate our capabilities in this regard, we have demonstrated a clear positive correlation between the beam divergence and the longitudinal confinement factor (see Figure 1). For this work, we consider a metric known as the D86 divergence angle – it’s the diameter of a circle that encircles exactly 86.5 percent of the laser’s total beam power (compared with the full-width at half-maximum, that’s a more stringent metric for evaluating small divergence). We have shown that we have precise control over the D86 divergence angle – spanning from 8° to 26° – across devices featuring 5 to 14 junctions.

Figure 2. Performance consistency across 500 wafers and typical intra-wafer divergence distribution. (a) D86 divergence angle of an AR-VCSEL product, evaluating of over 500, 6-inch wafers in production. (b) A typical D86 divergence angle distribution of the AR-VCSEL product in (a) within a 6-inch wafer.

Figure 3. Performance metrics for 8-junction and 9-junction arrays, including power density and efficiency. (a) Normalised light-current (L-I) curves for arrays with oxidation apertures (OA) ranging from 20 to 40 µm; current density is normalised to the total OA area, and array power density is normalised to the total array area (the minimal area to enclose all emitters). (b) Power conversion efficiency (PCE) as a function of current density. (c) Far-field divergence angle for 8-junction and 9-junction arrays across various OA sizes. (d) Temperature-dependent L-I characteristics for a 9-junction, 12-emitter array (30 µm OA) measured from 25 °C to 105 °C.

One of our key findings is that while the overall pattern of the laser array does not fundamentally alter the dependence of divergence on the confinement factor, the size of individual oxide apertures matters. This has led us to categorise our devices into small, medium, and large aperture groups. We note that divergence typically increases by 1° for every 5 μm increase in aperture size. Using this technology, we have realised a record-low full divergence angle of 8° and record-high brightness of approximately 140 kW mm^-2 sr^-1 in our multi-junction AR VCSELs. These performance figures meet the range and precision requirements for primary lidar.

Mass production success
For any device developer, one of the biggest challenges when moving a technology from a laboratory to mass production is ensuring extreme consistency. We have made much progress on this front, with our 6-junction AR VCSELs now produced in high volume.

Figure 4. Comparison of brightness (a) and spectral brightness (b) across four different epitaxial designs for 8-junction AR-VCSEL arrays with 42-emitters (OA 24 µm and emission area 225 µm by 218 µm).

According to performance data from more than 500 6-inch wafers, our manufacturing process has exceptional stability (see Figure 2(a)). The average wafer-median divergence angle across this massive sample size is 16.66°, and the standard deviation a mere 0.254°. This level of uniformity is essential for automotive manufacturers, as they require predictable performance across thousands of units. Intra-wafer distribution is also impressive, with a yield of nearly 100 percent against strict divergence specifications (see Figure 2 (b)).

Figure 5. Characteristics of a polarisation-controlled array (0.3 mm × 1.0 mm emission area, about 24 µm oxidation aperture) showing power and extinction ratio stability. (a) Pulsed light-current characteristics (20 kHz, 6 ns) at 50 °C; the inset displays the far-field profile at about 70 A with a D86 divergence of approximately 19 °. (b) Polarisation extinction ratio (PER) versus driving current under 50 °C pulsed operation. (c) Temperature sensitivity of the PER measured at about 70 A from 25 °C to 85 °C.

Based on all these results, we conclude that our AR VCSEL architecture is high-performing and highly manufacturable. We are transitioning this manufacturing excellence to AR VCSELs with 8 junctions that offer an even higher power density, while maintaining the same rigorous divergence standards.

Power scaling
An attractive option for increasing the power of the VCSEL is to add more junctions within the same vertical stack. With this approach, there’s a linear relationship between power and the number of epitaxial junctions.

Our AR VCSEL trajectory includes a rapid increase in junction count. Typically, we add a couple junctions every couple of years. Back in late 2021, 5-junction and 6-junction designs were industry standard, providing enough power for early hybrid scanning lidar systems. By 2024, 8-junction designs had emerged, providing a significant boost to brightness and enabling the first wave of long-range solid-state sensors. This year, architectures with 9 or more junctions are entering mass production.

We have successfully navigated the complexities of mass-producing 6-junction, 8-junction, and 9-junction AR VCSELs. With the introduction of every additional junction, we increase optical power and efficiency at a given current level. Our 9-junction arrays, featuring 30 μm apertures, are delivering an array power density of 5450 W mm^-2 that’s limited only by the maximum driving current of the driver (see Figure 3 (a)). Power-conversion efficiency of these VCSELs outperforms their 8-junction counterparts with various oxide aperture designs, especially under high driving currents.

However, careful engineering trade-offs must be made when scaling to higher power levels. Introducing more junctions increases the operating voltage, which must remain within the limits of automotive electronic drivers. While considering this matter, another ponderable is that larger aperture arrays generally offer higher fill factors by trimming the dead space between the emitters – but this tends to be realised alongside a slightly higher divergence (see Figure 3 (c)).

When balancing these various pros and cons, we often find that 30 μm devices represent the ‘sweet spot’ in design, outperforming 20 μm and 40 μm variants when trading power density against beam quality. In practical automotive applications, we tend to recommend maintaining current densities below 1000 A mm^-2 to ensure a sufficient lifetime margin.

Robustness, brightness and polarisation
Amongst the various applications involving deployment of semiconductor lasers, automotive environments are one of the most demanding. One issue is that ambient temperatures can fluctuate wildly.

We are encouraged by the results from testing of our-junction AR-VCSEL arrays at temperatures up to 105 °C. At this extreme, our emitters retain 80 percent of their room-temperature power (see Figure 3 (d)). Special designs could enable even more impressive high-temperature performance.

A multi-pronged engineering approach is behind the enhanced performance at over 100 °C. Success comes from focusing on minimising the thermal resistance of the package and the chip itself, while simultaneously optimising the gain spectrum of the quantum wells. We carefully engineer the degree of offset between the gain peak and the cavity resonance to ensure the laser remains in its most efficient operating regime, even as the temperature rises. By placing a premium on thermal robustness, we ensure lidar systems incorporating our VCSELs maintain a consistent detection range, regardless of the environmental conditions. That’s a prerequisite for Level 3 and Level 4 autonomous systems.

Figure 6. Addressable AR-VCSEL array configurations and performance. (a) Optical micrograph of a 1D addressable array comprising 64 individual sections. (b) Temperature-dependent light-current curves (25–105 °C) for a single section from the 1D array; the inset illustrates the far-field emission pattern during operation. (c) Optical micrograph of a 2D addressable array organized into 32 × 16 zones. (d) Light-current (L-I) characteristics across varying temperatures (25–105 °C) for an individual zone within the 2D array, with an inset showing the corresponding far-field profile.

We have also undertaken a series of controlled experiments to evaluate how variations in the anti-reflective region alter the confinement factor. We produced a family of devices with an 8-junction base epitaxial design. These AR VCSELs have divergence angles of 14.7°, 11.8°, 10.1°, and 8.1°. Compared with the design with a 14.7° angle, that with a value of 8.1° delivers a two-fold increase in brightness and a threefold increase in spectral brightness. As we transition these optimisations to our 9-junction platforms, we expect to push brightness levels higher.

Driven by the increasing sophistication of lidar systems, the focus for the optical source is shifting beyond raw power and towards signal quality. As part of this trend, polarisation control is emerging as a pivotal tool for increasing the signal-to-noise ratio in complex driving environments. When system designers utilise a light source with a fixed, stable polarisation, they can place a linear polariser at the receiver to filter out unpolarised ambient sunlight and glare from other vehicles. With this simple step, the signal-to-noise ratio can increase by approximately 3dB, effectively doubling sensitivity under certain conditions.

Beyond noise suppression, polarisation signatures offer a new dimension of data for material classification. Different road surfaces and objects – such as water, asphalt, vegetation, metal, or human clothing – depolarise light in unique ways. By analysing these changes, it’s possible for the autonomous vehicle to have a better understanding of its surroundings.

Based on this opportunity, we have developed polarised AR VCSELs with integrated gratings that produce a record-high polarised output of 630 W with a polarisation extinction ratio exceeding 12 dB. These polarised AR VCSELs deliver this level of performance across diverse operating conditions and temperatures (see Figure 5). While the integration of these gratings often results in a small power loss of 10 to 20 percent, our high-precision fabrication techniques minimise these effects, enabling a high-power density and stable polarisation across a wide range of operating currents and temperatures.

Shifting to all-solid-state lidar
For the automotive sensor, the ‘holy grail’ is an all-solid-state lidar. Such systems are highly desirable, combining compactness with affordability and the absence of mechanical moving parts, which are prone to wear and calibration issues.

To support this transition, we are developing and producing addressable AR-VCSEL arrays in both one-dimensional and two-dimensional formats. Two 8-junction addressable arrays are showcased in Figure 6: a 1D version (2 x 32 sections) with a 21° divergence ((a)-(b)), and a 2D version (16 x 32 sections) with a 24° divergence ((c)-(d)). Both chips maintain a footprint of approximately 5.0 mm x 3.8 mm.

While many of today’s addressable arrays are utilised for supplemental sensors, such as blind-spot monitors or short-range pedestrian detection, the future of primary long-range lidar lies in all-solid-state configurations. These systems will demand divergence angles significantly below 18° to maintain a high resolution over hundreds of meters. To meet this requirement, the AR VCSEL architecture is not just advantageous – it is an indispensable component. This form of emitter has a unique ability to provide low-divergence, high-power light in an addressable format, a set of strengths providing a technological foundation for the next decade of automotive sensing.

The road ahead
Thanks to high brightness AR-VCSELs, it’s possible to significantly extend the detection range and precision of VCSEL-based commercial lidar. This breakthrough may be widely adopted, as AR VCSELs provide a universal light source for all ranges of commercial lidar.

The introduction of AR VCSELs are not just elevating performance to a new level. This technological breakthrough will have a profound impact on the economics of lidar, as well as the electric vehicle market. By providing a high-performance light source, manufactured by leveraging existing high-volume semiconductor processes, AR VCSELs are helping to drive down the cost of primary lidar units towards the critical $100 mark. This is a tipping point, representing the threshold for mass adoption in mainstream passenger vehicles.

Results of greater affordability are already visible. Last year, lidar market penetration in the Chinese automotive sector exceeded 10 percent, with nearly all major manufacturers integrating these sensors into flagship and mid-range models. Over time, deployment will expand into basic models, with the overall penetration rate expected to reach 20-30 percent this year. And as AR-VCSEL technology continues to mature and scale, we forecast this trend to accelerate globally.