Light detection and ranging (LiDAR) technology, a laser-based imaging technique for accurate distance measurement, is considered as one of the most crucial technologies for autonomous vehicles, robots navigation, and aerial vehicle reconnaissance. Typically, LiDAR has relied on light sources and detectors mounted on mechanically rotating structures to cover the entire field of view [1]. However, advanced LiDAR technologies such as solid-state LiDAR that do not require mechanically rotating structures have emerged recently. Among these newly emerged technologies, solid-state LiDAR is commonly categorized into the following three types: flash-based LiDAR, microelectromechanical system (MEMS)-based LiDAR, and optical phased array (OPA)-based LiDAR.

Furthermore, advanced optical technology utilizes novel balanced photodetector (BPD) devices in LiDAR systems. LiDAR operates on the principles of three common schemes: pulsed time of flight (TOF), amplitude-modulated continuous wave TOF, and frequency-modulated continuous wave (FMCW) [2].

The FMCW technique has demonstrated broad potential applications. These include high-precision mapping, autonomous vehicle navigation, security surveillance, and underwater object detection [3–9]. FMCW detection technology can detect weak optical signals by employing the principle of coherent detection. In addition, combining the OPAs, FMCW ranging systems [10–13] are also proposed. The solid-state 3D-imaging system, realized by applying the Vernier OPA to the FMCW LiDAR system, has been previously reported [14].

Compared with the TOF [15,16] and phase modulation [17,18] approaches, FMCW-based LiDAR systems excel by enabling coherent detection. It offers several key advantages: superior accuracy and sensitivity, reduced optical power requirements, anti-ambient-interference, less stringent bandwidth demands on optoelectronic components [19], and detection of weak light. Additionally, the FMCW architecture demonstrates significant prospects for integration with compact silicon photonic chips [20,21], facilitating ultimate miniaturization and efficient sensing solutions.

To enhance the detection precision of LiDAR systems [13], a highly sensitive receiving device is required [22–24]. A single photodetector (PD) is easily affected by ambient light and other electromagnetic interferences, resulting in unstable measurement results. By detecting two opposite phase signals at the same time and subtracting them to reject common mode noise, the balanced detection technology can greatly improve the detection accuracy and stability. This technique is particularly suitable for applications where changes in weak signals need to be accurately measured.

A BPD with an on-chip biasing capacitor was designed and fabricated using CMOS-compatible silicon-on-insulator (SOI) technology by Hai et al. [25], achieving a common-mode rejection ratio (CMRR) of 30 dB. Houtsma et al. [26] demonstrated a high-power BPD with over 1 W of radio frequency (RF) output power at 2 GHz and OIP3 of 48 dBm at room temperature. Responsivity was 0.65 A/W, and CMRR was greater than 40 dB. Liu et al. demonstrated a four-channel dual-polarization FMCW LiDAR receiver module using a silicon photonic coherent receiver chip. The sensitivity of the module was tested to be better than −80dBm [23]. Typically, BPDs commonly adopt a two-electrode structure. However, their performance is restricted by the electrical architecture, especially the unidirectional current characteristics of two-electrode active devices [27] (such as p-n junction diodes). To break through these limitations, researchers have developed three-electrode Ge/Si avalanche photodiodes (APDs) [28,29]. Liu et al. [30] utilized an n-doped charge layer. By adjusting the voltage of the third electrode, they successfully reduced the dark current and applied this achievement to the field of imaging. Liu et al. [31] further optimized the internal electric field distribution by adjusting the voltage on the third Ge electrode, changing the potential differences between pGe-nSi and pSi-nSi. This method not only achieved effective control of the dark current but also obtained a high responsivity, making it suitable for weak light detection fields. Qu et al. [32] developed a n-type charge layer three-electrode Ge/Si APD. By optimizing the applied voltage, the device achieved the performance of low dark current and high bandwidth, which is particularly suitable for application scenarios with low voltage and high bandwidth.

In this work, a three-electrode balanced avalanche photodetector (3ele-BAPD) is proposed, inheriting the advantages of the traditional BPD. Furthermore, a third electrode is introduced to further optimize the CMRR and signal sensitivity. The third electrode is employed to manipulate the detector’s bias voltage, improving I–V characteristics and functioning as a reference electrode to optimize noise rejection. With this design, the 3ele-BAPD experimentally demonstrates exceptional performance in precision-critical applications, particularly excelling in weak light conditions and during imaging tasks. To further measure the weak light detection ability, the 3ele-BAPD experimental system based on FMCW was designed and studied. The frequency spectrum, along with the SNR and detection probability of the beat signal, was analyzed in detail. The two-electrode balanced photodiode (2ele-BPD) and two-electrode balanced avalanche photodiode (2ele-BAPD) were used for the performance references. In addition, the 3ele-BAPD was also applied to the FMCW system for 3D imaging and velocity measurement.

The proposed 3ele-BAPD structure is shown in Fig. 1(a). The 3ele-BAPD system takes advantage of silicon photonic integration, incorporating two germanium-on-silicon (Ge/Si) APDs and their corresponding optical waveguides onto a single chip. A single detector consists of two parts. One part is a traditional two-electrode Si-APD structure, as shown on the right side of Fig. 1(a). This Si-APD has low dark current and high gain. However, this part cannot respond to shortwave infrared (SWIR) light effectively. Therefore, the other part of the structure provides photo-induced charges, which are in the Ge absorption region. The Ge absorption region can be controlled by an independent electrode. The electric field in the Ge absorption region is mainly determined by the bias on the Ge electrode, while the electric field in the multiplication region of Si is controlled by the voltage applied between Si n++ and Si p++ electrodes. This feature facilitates the device responding to near-infrared light effectively, thus achieving a high responsivity and low dark current simultaneously.

The 3ele-BAPD devices were fabricated on SOI wafers with 220 nm top silicon at Advanced Micro Foundry (AMF), Singapore. Ge was grown on silicon by reduced pressure chemical vapor deposition (RPCVD) through a selective epitaxial growth process. After the device was completed, all the APDs were covered with silicon dioxide. The necessary fabricating steps for the device are as follows. The top silicon layer is etched to define the silicon grid pattern. Boron and phosphorus were implanted to form p-type and n-type regions, respectively. The Si p++ and Si n++ regions are the two terminals of the silicon. A Ge layer was then epitaxially grown on lightly doped p-type region [highlighted in pink in Fig. 1(b)]. The dimensions of the epitaxial Ge are 23μm(length)×4μm(width)×500nm(height). The Ge film is grown intrinsically and subsequently implanted on the top surface to form Ge p++ region, which creates the Ge termination contact. The Ge/Si 3ele-BAPD device schematic is shown in Figs. 1(a) and 1(b), which consists of a pair of APD1 and APD2 of the same structure. The specific electrode connection method can be seen in Figs. 1(b) and 1(c).

The I–V characteristics and responsivity were measured using the Santec TSL-510 laser and Keysight B1500A. The light was coupled into the silicon waveguide via a grating coupler and subsequently bent upward directly to couple into the germanium active layer of the detectors. Meanwhile, the probes contacted the APD’s electrode pads, which were connected to the B1500A for data reading. The measured responsivity was calculated by dividing the photocurrent with the input optical power.

Due to the complexity of 3ele-BAPD, the device can be used as the single 2ele-BAPD by floating one of the electrodes. The two-electrode configuration is to float the electrode on p-Si and apply reverse bias between E1-2 and E1-1 terminals. The photocurrent of the APD was measured as a function of incident light power at 1550 nm. As shown in Fig. 2(a), when the dark current is greater than 10 nA, the device will generate a significant photocurrent under low input optical power. Due to the increase of the electric field inside Ge, the photogenerated carriers in Ge are swept into the avalanche region in Si more rapidly to form a larger photocurrent. Figure 2(b) shows the responsivity of the two-electrode device under varying optical input powers (−10dBm to −50dBm) for different dark current levels. At a dark current of 10−9A, the responsivity remains nearly constant at ∼0.19A/W. As the dark current increases, the responsivity enhances significantly at lower optical powers: rising from 0.16 to 0.73 A/W (10−7A), from 0.16 to 2.63 A/W (10−6A), and from 0.34 to 33.60 A/W (10−5A).

In the three-electrode configuration, where E1-1 terminal is swept while E1-2 and E1-3 are grounded, the current and responsivity curves are shown in Figs. 2(c) and 2(d). At a dark current of 10−9A and 10−7A, the responsivity remains relatively constant at ∼0.87A/W and ∼1.74A/W, respectively. However, for higher dark currents, a pronounced enhancement is observed at lower optical powers: increasing from 2.40 A/W to 92.21 A/W at 10−6A, and from 2.38 A/W to 183.43 A/W at 10−5A, as the optical power decreases from −10dBm to −50dBm. Since the 3ele-BAPD realizes the carrier absorption and multiplication processes respectively by two pairs of independent applied voltages, the responsivity is further increased while reducing the dark current.

In our previous work [31], through theoretical TCAD modeling, the internal electric field distribution of a 3ele-BAPD was characterized. The analysis focused on the device’s response under a fixed voltage applied to the Ge electrode, with the Si p++ terminal grounded and the Si n++ terminal voltage swept. Designed to operate via an electron-initiated avalanche process, the device incorporates a bridging region [Fig. 1(b)] that spatially decouples the Ge absorption and Si multiplication regions. Simulation results reveal that increasing the Ge terminal voltage significantly increases the electric field within the Ge region. Concurrently, the electric field in the bridging region strengthens, facilitating efficient carrier transport into the Si multiplication region at near-saturation velocity. This design enables dynamic control over the Ge absorption region’s electric field without sacrificing the high field strength required in the multiplication region, offering a promising pathway to optimize responsivity and dark current simultaneously.

To study the operating voltage of the 3ele-BAPD, light is introduced via the grating couplers at both terminals to measure the photocurrent and dark current separately of the two single APDs. The current flowing into a device is defined as positive, while the current flowing out of the device is defined as negative. The voltage application method for APD1 involves scanning the E1-1 terminal while applying 0 V to both the E1-2 and E1-3 electrodes. The light-dark current curves and responsivity are shown in Figs. 3(a) and 3(b). When the dark current is 10 μA, the optical power decreases from −10dBm to −50dBm, and the responsivity increases from 2.16 A/W to 115.12 A/W. In the linear region of the I–V curve, under an operating voltage biased at 14 V and an incident optical power of −10dBm, the photocurrent flowing into the E1-1 terminal is 63 μA. To ensure a positive photocurrent flows into the E2-1 terminal of APD2, the voltage application method for APD2 is designed to synchronously scan the E2-2 and E2-3 terminals while maintaining 0 V at the E2-1 terminal. The light-dark current curves and responsivity are shown in Figs. 3(c) and 3(d). When the dark current is 10 μA, the optical power decreases from −10dBm to −50dBm, and the responsivity increases from 1.20 A/W to 200.54 A/W. In the linear region of the I–V curve, under an operating voltage biased at −14V and an incident optical power of −10dBm, the photocurrent flowing into the n-Si terminal is 63 μA. The operating procedure novelty of the BAPD in this work is the connection of the E1-3 region of APD1 and the E2-1 region of APD2, thereby rejecting the common mode signals of APD1 and APD2.

The BPD uses a pair of photodetectors to perform photoelectric conversion on two optical signals, respectively, and amplifies the difference of two electrical signals obtained, thereby achieving the purpose of enhancing the differential mode signal and suppressing the common mode signal. In practical measurement, the common mode rejection ratio is widely adopted to assess the consistency and symmetry of a balanced detector pair. Superior structural symmetry and uniformity of the device contribute to enhanced performance. Commercial applications typically require a CMRR exceeding 25 dB for optimal operation. A higher CMRR reduces noise and improves the SNR, facilitating clearer and more accurate signal detection in photodetector systems [33].

The CMRR can be defined as [23] CMRR=−20×log(|I1+I2||I1−I2|).(1)I1 and I2 represent the direct photocurrent at the n-Si terminal of APD1 and APD2, respectively. If the optical power injected into the APDs is the same, the static CMRR will be determined by the difference in responsivity alone. Under the operating voltage, the CMRR of the conventional 2ele-BPD and 3ele-BAPD was measured at 500 ms intervals within the wavelength ranges of 1549 nm and 1551 nm, respectively, as shown in Figs. 4(a) and 4(b). The results demonstrate that at the wavelength of 1550 nm, the 3ele-BAPD achieves an average CMRR of 50 dB, which is higher than that of the 2ele-BPD (35 dB). This represents a 15 dB improvement in direct current component rejection over the conventional 2ele-BPD. This is because, compared to conventional PD, the APD incorporates an internal gain mechanism that generates measurable photocurrent even under weak light conditions, thus delivering superior sensitivity. By harnessing the responsivity characteristics of APDs within a balanced detection architecture, we achieved a superior CMRR over standard BPD. Furthermore, when benchmarked against commercial Newport InGaAs detectors (50 dB CMRR, 125 kHz bandwidth), our Ge/Si 3ele-BAPD achieves comparable CMRR while delivering substantially higher bandwidth, demonstrating its strong competitiveness and promising application potential for balanced photodetection.

In particular applications, FMCW LiDAR for instance, it is necessary to ensure that the sensitivity of the detector changes hardly or only slightly with the frequency. This is because the unevenness of the frequency response will increase the instability of the FMCW LiDAR system. The S parameters of the device were measured by a vector network analyzer (VNA) to analyze the RF characteristics and compare the frequency response of the devices. The VNA acts as both signal generator and receiver, with the receiver synchronized to capture signals at the same frequency. The laser source, modulated by the VNA via a Mach–Zehnder modulator (MZM), serves as the optical transmitter. A variable optical attenuator (VOA) is used at the laser output to regulate the input power and avoid detector saturation. For frequency response testing, the modulated optical signal is directed through a 2×2 coupler and then directed to the detector for photoelectric conversion, enabling evaluation of the individual responses from the BPD and BAPD.

The normalized bandwidth of the traditional 2ele-BPD was measured at a bias of −1V, and the optimal bandwidth was 1.15 GHz as shown in Fig. 5(a). Figure 5(c) is the normalized bandwidth of the 3ele-BAPD. At a bias of −55V, the optimal bandwidth is 340 MHz, which adequately meets the requirements of most automotive LiDAR applications. One of the advantages of FMCW technology is the relatively low required sampling frequency even if the modulation bandwidth is relatively high [24,34], which is a consequence of the inherent homodyne configuration of the LiDAR front-end [35,36]. To achieve higher speeds, the bandwidth can be further increased by reducing the capacitance, which in turn can be accomplished by decreasing the area of the epitaxial Ge film.

It is worth noting that the dynamic CMRR is influenced not only by the amplitude of the photocurrent but also by the frequency response of the APD. The CMRR can be measured as a function of frequency. Specifically, at a given frequency, the CMRR is determined by the difference between the RF response when only one PD output is connected and the common mode or balanced mode differential signal obtained when two PDs operate simultaneously at the same frequency. In the frequency response measurement experiment, we maintained a consistent experimental configuration while sequentially replacing the devices under test (DUTs). This approach enabled us to separately characterize the common mode and differential mode responses of both the BPD and BAPD.

The RF response curves of the differential and common modes for the 2ele-BPD and 3ele-BAPD were measured, and their corresponding CMRRs were derived, as shown in Figs. 5(b) and 5(d), respectively. The comparison reveals distinct differences in the differential mode signals between the two detectors, as the two chips employ different detectors and exhibit distinct response characteristics. Notably, the common mode RF response of the 3ele-BAPD is significantly lower than that of the 2ele-BPD. By extracting the differential mode and common mode responses within the frequency range below 100 MHz, the CMRR performance of the device under actual operating bandwidth conditions can be effectively assessed. Operating at a 1550 nm incident wavelength, the 2ele-BPD achieves a CMRR of 27 dB. In contrast, the 3ele-BAPD attains a CMRR of 44 dB, representing a significant improvement of 17 dB in the low-frequency range. The CMRR improvement in the 3ele-BAPD benefits from its superior frequency characteristics, which are influenced by carrier transit time, RC time constant, and parasitic elements [37].

A distinguishing feature of FMCW LiDAR, which sets it apart from other LiDAR systems, is its coherent detection of the reflected signal. It enables the measurement of ranges and velocities of multiple targets simultaneously. In particular, a frequency-swept laser source deployed in the FMCW LiDAR system is used to measure the distance of a target via an RF beat signal, which is generated by the coherent combination of the local reference signal and the reflected signal, followed by photoelectric conversion through a BPD.

Achieving a lower minimum detectable power is essential for the successful implementation of both fiber optic long-distance links and free-space optical links, as it enhances signal sensitivity and overall system performance. We designed an experimental device to indirectly estimate the detectable optical power limit of the FMCW system, as shown in Fig. 6(c). The frequency-modulated signal is divided into signal light and reference light through a 98:2 beam splitter. The signal light is connected to a VOA to adjust the optical power, and then a 20 m long optical fiber is used to replace the free space transmission distance. Since the detectable optical power limit is very low, it cannot be directly measured by a power meter. In consequence, after passing through the 20 m optical fiber, the light is evenly divided into two channels through a 50:50 beam splitter. The power (Pm) of one channel is measured by an optical power meter; the other channel is connected to a 40 dB fixed optical attenuator (FOA) to further attenuate the signal light power and finally is mixed with the reference light signal; then the optical signal is converted into an electrical signal by a BPD. The signal optical power can be expressed as Preceive=Pm/10000. The detectable power limit of the optical power meter is −70dBm, so this device can accurately control the receiving power to −110dBm. The spectrum data are collected and analyzed by the host computer program.

Figure 6(a) presents the frequency spectra of beat signals obtained from the 2ele-BPD under received optical powers ranging from −64.5dBm to −89.5dBm. As the input power decreases, the beat signal amplitude diminishes accordingly. When the received power is −64.5dBm, the beat signal power is −75.4dBm; at −86.5dBm, it drops to −96.5dBm, corresponding to an SNR of 2.9 dB. Though beat signals can still be discerned at −89.5dBm, the risk of noise overlap increases significantly. Figure 6(b) shows the spectra for the 3ele-BAPD with received powers from −75dBm to −94dBm. The beat signal power decreases from −82.4dBm at −75dBm to −103.5dBm at −93dBm, where the SNR remains at 3.2 dB, demonstrating effective detection of ultra-weak optical signals.

First, the 3ele-BAPD LiDAR system is applied to measure velocity as shown in Fig. 8. The bandwidth for the linear frequency excursion is 3 GHz, and the period for the up-chirp and down-chirp time is 100 μs. The coherent LiDAR can directly measure the target velocity through Doppler frequency shifts of the received light [4]. The velocity-annotated 4D point cloud image of a rotating wheel is illustrated in Fig. 8(a). Figure 8(b) shows the measured beat signals for a static wheel and a rotating wheel. When the wheel stops, the beat signal of the up-chirp is consistent with that of down-chirp, indicating the measured frequency is 3.9 MHz. When the wheel starts rotating, the up-chirp and down-chirp signals shift in the opposite direction, and the median of the two signals is the frequency of 3.9 MHz; their difference can be solved as the velocity of −0.522m/s. We utilize the stepping motor to steer the light beam along the edge of the rotating wheel. The distance of 21 positions and corresponding projection of velocity in the direction of beam propagation are measured, as shown in Fig. 8(c). The velocity projection along the edge of the rotating wheel varies linearly from −0.6 to 0.6 m/s. It can also be found that there is a jitter in the outline of the wheel in the point clouds. It may be attributed to the instability of the system caused by the nonlinearity of chirp modulation and the linewidth of the laser [34,38].

Furthermore, we conducted a 3D imaging experiment of LiDAR based on stepping motor. As shown in Fig. 9, it is a high-resolution 3D point cloud image with more than 7600 resolvable points, including 201 horizontally resolvable points and 41 vertically resolvable points. The inset shows the corresponding camera image of the real scene. The 3D point clouds clearly exhibit a series of Chinese characters located at the middle area of the field of view (FoV) with a white background. The stepping motor has flexible angular resolution, which can provide more detailed 3D images by increasing the angular resolution. Complex strokes require extremely high angular resolution scanning. For strokes with simple outlines, such as a thicker horizontal line, moderate resolution can also clearly display the outlines. Therefore, the horizontal area of Chinese characters is scanned with a 0.1° angle resolution to better display details, while the simple vertical outlines are scanned with a 0.2° angle resolution. The outline of the Chinese characters can be clearly distinguished, and the shadows of the Chinese characters on the back wall are also clearly presented. The point clouds of the background are also specially displayed. The gradual change in distance from ∼2.2 to 4.5 m can be accurately observed through the changes in colors. By virtue of the FMCW coherent detection supported by the proposed three-terminal device, the 3ele-BAPD-based LiDAR exhibits superior and uniform quality of 3D imaging over the whole FoV. To our best knowledge, this is the first time that 3ele-BAPD LiDAR has demonstrated 3D imaging. The stepping motor-based FMCW LiDAR system can potentially process more than the demonstrated 7600 points in 3D imaging.

This work demonstrates a three-electrode Ge/Si BAPD for FMCW LiDAR application. A pair of 3ele-BAPDs and optical waveguides are integrated on chip to create a LiDAR receiver system. The operating voltage and frequency spectrum are measured, yielding the minimum detectable power. The results from comparative experiments between the 3ele-BAPD and 2ele-BPD demonstrate that the 3ele-BAPD, as an emerging solution, offers significant advantages in environments with reduced light levels. The proposed 3ele-BAPD increases the CMRR by 15−17dB at the wavelength of 1550 nm. Owing to the significantly enhanced CMRR of the 3ele-BAPD, LiDAR systems can effectively acquire FMCW-based beat signals while simultaneously measuring the SNR and detection probability of the signal light. Compared to the two-electrode experimental setup, statistical spectral analysis shows that the three-electrode setup achieves a significantly lower minimum detectable power of −93dBm. Consequently, the advantages of the 3ele-BAPD method become even more apparent. To further optimize device performance, integrated optimization strategies may be implemented. The splitting ratio of the 2×2 coupler and bandwidth characteristics are crucial for improving the CMRR. In the future, we will design a 2×2 coupler with an optimized splitting ratio and higher bandwidth, and integrate it with the 3ele-BAPD.

In addition, the novel 3ele-BAPD receiver can be manufactured with high yield using a standard CMOS process. The different performance indicators of this APD, such as breakdown voltage and responsivity, can be achieved on the same wafer. The 3ele-BAPD has broken the inherent barrier of the previous BPD device. This is the first and a major breakthrough. To the best of our knowledge, it is the first time that 3D imaging has been realized based on the 3ele-BAPD LiDAR system.

In summary, the proposed 3ele-BAPD can improve the detection of weak light signals. For LiDAR systems, it provides higher sensitivity and weak light detection capabilities, which are crucial for applications such as self-driving cars, thus showing significant practical value.