Solid-State LiDAR With Optimized Dynamic Beam Power Intensity for Reduced Thermal Load

Precise sensing of moving vehicles and stationary three-dimensional (3D) objects is enabled by light detection and ranging (LiDAR) systems. LiDAR sensors produce distance and velocity information about various targets by scanning the surrounding environment with an optical beam. A $360$ deg rotating mechanism is often used in a standard fashion to achieve this sensor scan function [1]. Such sensors may be bulky due to mechanical housings, bearings, electrical wiring, internal electronics, and associated heat sinks. The miniaturization of LiDAR sensors is thus attractive since a smaller electronics package size allows for tighter integration with autonomous land-based and aerial vehicles for a range of future mobility applications. Nonetheless, the down-sizing of a traditional LiDAR sensor package presents a host of challenges due to the aforementioned rotating mechanical parts. Over the last decade, significant resources have therefore been committed toward the research and development of solid-state LiDAR systems [2].

As reviewed by Li et al. [2], solid-state LiDAR systems may utilize pulsed or continuous wave sensing schemes. Pulsed time-of-flight (TOF) systems have a transmitter (TX) that emits an optical pulse that is then reflected off the sensed object back toward a receiver (RX). Such pulsed systems have high peak power with low average power to meet longer sensing distance and eye safety requirements [3], respectively. Amplitude-modulated continuous wave TOF systems are better suited for moderate range sensing due to the use of intensity modulated light with a phase difference between TX and RX. In contrast, frequency-modulated continuous wave (FMCW) systems collect a received optical signal from a transmitted frequency-modulated signal. The FMCW method is attractive due to the benefits of reduced sensor-to-sensor interference, Doppler effect informed velocity measurement, and reception of multi-target returns from a single sweep, in general. However, compared with pulsed TOF systems with relatively low average power, FMCW systems have higher average power owing to their reliance on a 100% duty cycle, resulting in an increased thermal load to the electronics package from the light source or laser.

From a device packaging perspective, solid-state LiDAR is a highly promising technology for sensing in future autonomous vehicle systems. A silicon photonics chip-scale device approach enables a lower cost, higher reliability solution, where all mechanically moving parts are eliminated. Optical phased arrays using FMCW sensing schemes for TX and RX have been demonstrated for collimation and steering of the beam for range measurements [4,5]. A representative schematic of a chip-scale LiDAR device with a zoomed view of basic construction is shown in Fig. 1. To generate the laser input for the beam in a chip-scale package, a light source (i.e., laser diode or gain chip) is typically employed, which generates heat that is easily managed in a controlled laboratory environment and when the light source itself is positioned off-chip. Integration of the light source on to a chip that is then placed in a harsh vehicle environment presents additional thermal challenges for not only LiDAR but also other integrated photonics systems. As discussed in Refs. [6–8], on-chip integrated lasers have several challenges including poor thermophysical material properties of typical silicon-on-insulator mounting substrates, junction-side-up architectures that make device heat load removal difficult, and different maximum temperature requirements for standard electronics versus photonics platforms. Thus, while solid-state LiDAR with a FMCW sensing scheme shows great potential, solutions to thermal challenges related to light source temperature control are nonetheless required.

To address the above challenges, a range of cooling solutions are available for optoelectronic packages, as shown in Fig. 2. An air-cooled heat sink in combination with a macro-thermoelectric cooler (TEC) is a standard thermal management approach for laser diodes, Fig. 2(a), although numerous package layers and thermal interface materials (TIMs) with associated conductive thermal resistances are of concern. Alternatively, aggressive thermal solutions that utilize micro-thermoelectrics (⁠$μ$TECs) [6] (Fig. 2(b)), vapor chambers [9] (Fig. 2(c)), or liquid flow through microchannels (⁠$μ$ channels) in a substrate that is directly attached to the device [6,10,11] (Fig. 2(d)) for near-junction laser cooling have been proposed. Here, advantages include fewer thermal resistances (i.e., structural layers and TIMs) between the heat source and sink in the photonics package leading to precise temperature control, while drawbacks may include higher cost plus greater integration and cooling system complexity.

Another potential method to regulate the temperature of on-chip light sources is the minimization of the laser output power, although as explained above, higher peak power often equates to longer detection distances which are advantageous. Thus, historically and importantly, the transmit power may only be limited by eye safety considerations [3]. Additionally, as discussed in Refs. [1,12], when using traditional mechanically rotating LiDAR, scanning might occur at a fixed frequency along predetermined horizontal lines, and dynamic control of the beam intensity may provide limited benefits. By comparison, solid-state LiDAR with optical phased arrays provides opportunities for programable scanning and scheduling similar to well-developed radar systems [13]. Specifically, for LiDAR, electronic control of the beam with optical phased arrays allows for a similar quality of depth estimation to be made as traditional mechanically scanned sensors with a significantly reduced number of range measurements [12]. Regions for scanning may be selected based on targets of interest, terrain, recent or concurrent observations, sampling density, or energy-based priorities [14,15]. Furthermore, for 3D single-photon LiDAR imaging [16], programable scanning offers notable benefits in terms of increased scanning speed and reduced data volumes.

In this work, we contribute a unique method for programable scanning and beam intensity control of solid-state LiDAR that enhances the thermal performance of an integrated light source. The work is multidisciplinary since it lies at the intersection of the fields of optical phased arrays, photonics packaging, and device/sensor thermal management. The article is organized as follows. Important regions of a representative optoelectronics package for thermal modeling are first explained in Sec. 2. From there, we describe the thermal modeling strategy (Sec. 3), which is based on a proposed beam power modeling and scan optimization approach for optical phased arrays (Sec. 4). The impact of the programable scanning technique on the performance of the subject optoelectronics package configuration is then shown in Sec. 5. Here, thermal performance benefits are illustrated in terms of maximum temperature reduction for the optoelectronics package. The results show how the approach may be logically applied to LiDAR sensors in a land-based vehicle environment and how the cooling solution may be simplified. Conclusions are provided in Sec. 6.

Consider the representative LiDAR optoelectronics package schematic illustrated in Fig. 3. This heterogeneous package consists of several low power integrated circuit (IC) electronic devices soldered to a printed circuit board (PCB) and higher power photonic devices (i.e., quantum wells on an indium phosphide (InP) substrate with gold (Au) and silver (Ag) metallization) placed on a separate aluminum nitride (AlN) substrate along with the silicon (Si) LiDAR antenna array region. Thus, the package comprises three main zones for consideration in terms of thermal design, as shown in the zoomed cross-sectional views in Fig. 3. Typically, the upper surface of the package shown in this figure is outward facing (with the light source junction-side-up) to the environment for scanning purposes. Thus, a lidded package for heat sink attachment and thermal management is not realistic, and the cooling solution is appropriately shown at the bottom of the package in each cross-sectional image in Fig. 3.

To understand the limitations of integrated optoelectronics in the context of the package explained above, one-dimensional (1D) thermal resistance network models were developed along three heat flow paths corresponding to electrical and optical devices (Fig. 4). The three thermal resistance networks consider the heat flow paths starting from the heat sources and ending at the solid metal baseplate of the heat sink. Convective resistance of the heat sink is omitted, as it varies based on the selected cooling solution. The conduction thermal resistance of a given layer, $Rth_cnd$⁠, is a function of the thickness of the layer, $t$⁠, the effective heat flow area, $Ae$⁠, and thermal conductivity, $k$⁠, per the equation, $Rth_cnd=t/Aek$⁠. To simplify analysis, heat flow is assumed to propagate at a $45$ deg angle through each layer, thus increasing the effective heat flow area, $Ae$⁠, in each subsequent layer.

For the IC electronics device heat flow path (Fig. 4(a)), cooling is limited by the PCB and grease TIM layers. Increasing the number of through-plane thermal vias in the PCB, alternative electrical isolation, or improved TIMs may be used to reduce the total conductive thermal resistance from the electronic device to the heat sink. Under the gain chip (Fig. 4(b)), there is a set of large conductive thermal resistance layers including a thin silicon dioxide layer and the silicon substrate. Embedding liquid cooling channels in the AlN substrate, using enhanced external cooling such as a TEC, or implementing highly conductive heat spreading by way of a vapor chamber (Figs. 2(a), 2(c), and 2(d)) do not directly address the large conductive thermal resistances, which diminish the effectiveness of these cooling solutions. Bypassing this thermal resistance using a $μTEC$ embedded on the top of the light source gain chip [6] proves the most effective cooling mechanism; however, this solution comes with increased cost and manufacturing complexity. This motivates the exploration of alternative strategies for simplified package thermal management. While the total conductive thermal resistance from the gain chip to cooler is relatively high, the total conductive thermal resistance from the antenna to the heat sink is relatively low and limited by the grease layer; see the lower right bar chart of Fig. 4(c). Given the package resistance is relatively small, heat from the nearby gain chips play a large role in the performance of the antenna. While the 1D thermal resistance model provides a rough estimate of challenges for effective cooling, it does not fully consider lateral heat spreading which becomes prominent in such integrated packages, where temperature sensitive electronic devices may be near high heat flux photonic devices.

A mesh convergence study using steady-state analysis was performed by sweeping the number of finite elements between $6.2×104$ and $3.32×106$⁠. A mesh with $1.5×5$ elements was chosen to conduct all simulations to reduce computational expense, and a relative error of $<4%$ compared to the finest mesh size can be expected.

The temperature distribution in the model assuming constant gain chip and IC power loss is shown in Fig. 5(b), and a maximum temperature of 116 $∘$C is observed at the heat source of the gain chip (Fig. 5(c)). This simulation result confirms the large conductive thermal resistance of the package given the large temperature gradient below the gain chip and provides a reference for the proposed power reduction strategies that follow. Thus, the following studies consider the use of an optimized scan power profile, which is then input to the light source domain, $ΩGAIN$⁠, as a transient thermal load.

As motivated in the introduction, solid-state LiDAR with optical phased arrays provides opportunities for programable beam scanning and scheduling at shorter detection distances when maximum allowable beam power is not necessary. The concept for LiDAR with time-varying output power versus a constant power scheme is shown in Fig. 6. In the following sections, the LiDAR system is described, implementation of the time-varying beam output power modulation approach is explained, and scenario implementation is outlined.

We assume a representative solid-state scanning LiDAR similar to Ref. [18] which has a fast-axis controlled entirely by phase shifters, used to scan in azimuth, and a slow-axis where scanning in elevation is controlled by modulation of the laser output wavelength due to the use of grating antennas along that axis. Phase shifters allow the beam to be repositioned nearly instantaneously between two points which allows for dynamic on the fly programing of beam locations, but wavelength scanning is typically a linear process, without the possibility of beam hopping due to inherent performance limitations of the laser diode. Therefore, we constrain ourselves to reorder temporal activation of beam locations in only the azimuth scan direction. We set the system’s azimuth field-of-view to $160$ deg with 800 points per scan line, and set an elevation field-of-view to $37$ deg with 90 scan lines and a refresh rate of 10 Hz.

To reduce the thermal impact of a continuously operating laser on the LiDAR package, we reduce the laser power where possible to in turn reduce the thermal load. Reducing the power is accomplished through two main methods, the first by applying a five-sided bounding box around the scan area that delineates the maximum range we desire in all directions and the second by reducing power in directions that contain nearby targets in which case full power is not necessary to achieve a sufficient signal-to-noise ratio (SNR). In the first case, the bounding box size is determined by the type of road being driven. Highways with no cross streets allow for a narrow bounding box as constant wide angle scanning generally becomes unimportant and high speeds require long down-road scanning. Conversely, in slow speed urban environments, wide angle targets become more important relative to down-road targets. Additionally, scan lines with large angle in elevation typically have shorter distances of interest due to the scan line intersecting the ground or traveling upwards into the sky. In the second case, the previous LiDAR point cloud frame can be used to determine which pixels will contain a nearby target with high likelihood and appropriately adjust the required transmit power.

Often a situation arises where one beam that requires full power is spatially adjacent to a beam which can have reduced power. Each beam has a short dwell time making it unreasonable to directly modulate the laser power within such a time scale. Therefore, beams on each scan line are reordered based on optical power requirements rather than azimuthal position. Each scan line’s power loss, $Q(τ)$⁠, is concatenated together in time to give the power profile for a full scan frame, and we apply a low-pass filter with roll-off at 10 kHz to limit transient power changes to a reasonable slew rate for a direct drive scheme. Finally, some data reduction is performed to reduce the number of time-steps in the transient thermal simulation.

To obtain a representative point cloud, raytracing in a 3D scene is done via cuboid simulation provided by matlab’s Automated Driving Toolbox [21]. A simulated LiDAR sensor is attached to an ego vehicle which traverses the scene while outputting point cloud data at a chosen update rate interval. A bounding box, as defined in Sec. 4.1, surrounds and moves with the ego vehicle so that the distance to the bounding box for each beam can be determined, which is used to calculate the power limit, as described in Sec. 4.2. Scenarios can be populated with other dynamically moving road users, but in practice the ratio of points being returned by road users versus the surrounding buildings or bounding box is small for the purpose of device thermal modulation.

To evaluate the impact of the proposed programable beam scanning algorithm on the optoelectronics package thermal performance, the prior sets of simulation tools are combined, as shown in Fig. 7. This includes an optical simulation to determine beam power, optimizing and filtering the beam power profile, and finding the package temperature through a series of thermal simulations.

Two scenes were selected to demonstrate the proposed beam optimization strategy, one highway scene (Figs. 8(a)–8(c)) and one urban scene (Figs. 8(d)–8(f)). The highway scene consists of an open road with several vehicles and some far-away buildings as shown in Fig. 8(a). A bounding box that is 60 m wide, 20 m tall, and open ended in length is applied to limit the scan power. The spatially varying power profile for the optimized scan is shown in Fig. 8(b), where a majority of the scene consists of low power scans due to the open nature of the scene. The corresponding transient power loss profile, $Q(τ)$⁠, is provided in Fig. 8(c). This optimized highway scene has a total root-mean-square (RMS) power of 3.46 W compared to the 8 W total continuous power for two gain chips. For reference, the urban scene concatenated scan power profile is shown in Fig. 9 after low-pass filtering and prior to data reduction. Observe that the optimized scan retains its shape after application of the filter.

The urban scene is a representative city block layout that includes an empty intersection with several buildings and trees (Fig. 8,$(d)$⁠). A bounding box that is 240 m wide, 20 m tall, and 120 m long is applied to limit the scan power. The spatially varying power profile is shown in Fig. 8,$(e)$⁠, where a majority of the scene consists of low power scanning due to the proximity of the buildings and structures on the street. This observation is supported by the transient power loss profile, $Q(τ)$⁠, in Fig. 8(f). This optimized urban scene has a RMS power of 3.57 W, again compared to the 8 W total continuous power for two gain chips.

Given the complexity of each 0.1 s time-dependent power loss profile in Figs. 8(c) and 8(f), running the transient thermal simulation while cycling repeatedly through one of these profiles is not computationally efficient. In lieu of this, we first estimate an initial condition for the temperature of the package using a piece-wise linear RMS power profile approximation that was constructed for each of the two optimized power profiles, see Figs. 10(a) and 10(b). These surrogate profiles were then repeated 10 times in each thermal simulation (using a fixed 0.01 s time-step for resolving larger time scale transient events) to provide a longer 1 s input load profile. Full transient simulations were then executed (using an adaptive time stepping scheme with a maximum time-step size of 0.01 s to resolve finer transient features) to reach an approximately steady-state thermal response for each package (Fig. 10(c). Note that the device and package temperature distributions from each of the surrogate models after 1 s was used as an initial condition, and the full transient power profile consisting of 617 input data points for the highway scene and 684 points for the urban scene was simulated for each case (Fig. 10(d)).

In Fig. 10(d), it is clear from the constant power curve that the system has not yet reached a steady state, as the maximum device temperature rises from $107.2∘C$ to $108.2∘C$⁠. The system would take another 29 s to reach steady state, where a maximum device temperature of $116∘C$ is observed. For the urban scene, the maximum device temperature starts at $55.8∘C$⁠, reaches a peak value of $95.8∘C$⁠, before cooling to $53.5∘C$⁠. This suggests that the system peak temperature is an overestimate given that the final temperature is lower than the initial temperature. A similar trend is observed for the highway scene, where the maximum device temperature starts at $61.2∘C$⁠, reaches a peak of $99.8∘C$⁠, before cooling to a maximum of $56.4∘C$⁠. These trends suggest a 23% and 19% reduction in maximum device temperature rise (from $30∘C$⁠) during steady-state operation for a given scene. The $∼20%$ reduction in temperature rise for either case is a result of minimizing the number of times full power is used to scan identified areas of interest in the scene. While the urban scene had a higher RMS power, the peak temperature is slightly lower due to the scan profile, where the highway scene has more consistent full power scans thus increasing the package temperature.

The use of a 3D model also enables the evaluation of lateral heat spreading effects. The temperature of the ICs and antennas under three operating conditions are shown in Fig. 11. The peak temperature in the antenna region drops from $42∘C$ to $31.9∘C$ when optimizing the scan profile. Furthermore, the spatial temperature variation across the antenna surface is reduced from $5.7∘C$ to $1.7∘C$⁠. A similar trend can be seen across the ICs, where the peak temperature drops from $40.4∘C$ to $33.1∘C$ and the temperature range across the devices is reduced from $2∘C$ to $0.7∘C$⁠. A reduction in the temperature variation is desirable as both photonic and electronic devices tend to be highly sensitive to such metrics, especially optical waveguides in optical phased array LiDAR where refractive index is highly dependent on temperature and the signal phase delivered to each antenna must be exactly controlled. Thus, the programable beam scanning strategy effectively reduces the spatial temperature variation independent of scene selection.

In this research, programable beam scanning was proposed to reduce thermal management requirements for a solid state, chip-scale phased array LiDAR package. One-dimensional thermal modeling was used to understand the total thermal resistance for both electrical and optical components, and package conductive thermal resistance was identified as the key bottleneck to improve thermal performance. This motivated the use of beam scan optimization to reduce the transient scan power and thus time-dependent thermal load. A highway and urban scene were selected to demonstrate the method, and the optimized beam scan profile reduced RMS power by 50% in both cases. Three-dimensional transient thermal finite element simulations were then solved to understand the change in peak package temperature realizable using the optimized scan profiles. A $∼20%$ reduction in maximum package temperature rise at the light source was found for both scenes while still providing full angular coverage, thus verifying the effectiveness of the proposed approach.

Customized strategies such as beam decimation based on previous frame scans augmented by artificial intelligence algorithms trained on likely target locations are left as topics for future work. Such strategies might allow for increased dwell time of maximum range beams, further enabling reduced laser power and possibly lowering peak package temperature. Optoelectonics packaging innovations to reduce conductive thermal resistance between heat source and sink, considering such challenging sensor applications, is another important area for further development.

This study was supported by MIRISE Technologies.

This article does not include research in which human participants were involved. Informed consent not applicable. This article does not include any research in which animal participants were involved.

The datasets generated and supporting the findings of this article are obtainable from the corresponding author upon reasonable request.

$h$ =

heat transfer coefficient (⁠$Wm−2K−1$⁠)

$k$ =

thermal conductivity (⁠$Wm−1K−1$⁠)

$t$ =

thickness (mm)

$A$ =

area (⁠$m2$⁠)

$D$ =

aperture (m)

$I$ =

current (A)

$N$ =

noise power (W)

$P$ =

power (W)

$Q$ =

volumetric power density (⁠$Wm−3$⁠)

$R$ =

range (m)

$T$ =

temperature (⁠$∘C$⁠)

$V$ =

voltage (V)

$Cp$ =

specific heat capacity (⁠$Jkg−1K−1$⁠)

$Rth$ =

thermal resistance (⁠$KW−1$⁠)

$q″$ =

heat flux (⁠$Wm−2$⁠)

$ε$ =

efficiency

$η$ =

transmission factor

$θ$ =

incident radiation backscatter angle (deg)

$ρ$ =

density (⁠$kgm−3$⁠)

$σ$ =

reflectance

$τ$ =

time (s)

atm =

atmosphere

cnd =

conduction

cnv =

convection

$e$ =

effective

$l$ =

laser

$L$ =

layer

$RX$ =

receiver

sys =

system

$t$ =

target

$TX$ =

transmitter

1D =

one-dimensional

3D =

three-dimensional

AMCW =

amplitude-modulated continuous wave

CW =

continuous wave

FMCW =

frequency-modulated continuous wave

IC =

integrated circuit

LiDAR =

light detection and ranging

PCB =

printed circuit board

RMS =

root-mean-square

RX =

receiver

SNR =

signal-to-noise ratio

TEC =

thermoelectric cooler

TIM =

thermal interface material

TOF =

time-of-flight

TX =

transmitter