(2 中国科学院大学 材料与光电研究中心, 北京 100049)
(3 中国科学院脑科学与智能技术卓越创新中心, 北京 100083)
(2 Center of Materials Science and Optoelectronics Engineering, University of Chinese Academy of Sciences, Beijing, 100049, China)
(3 Center for Excellence in Brain Science and Intelligence Technology, Chinese Academy of Sciences, Beijing, 100083, China)
The low-cost Time-Of-Flight (TOF) sensor is increasingly becoming attractive in varied applications such as face recognition, 3-D gaming, autonomous driving and Advanced Driver-Assistance Systems (ADAS) . Single Photon Avalanche Diode (SPAD) based TOF sensor is attractive due to its ability to detect low intensity incoming light and to sense the arrival time of the photons with a time resolution of hundreds of picoseconds . Moreover, since the SPAD behaves almost like digital devices, processing can be greatly simplified .
The integration of SPADs in CMOS technology has significantly improved the level of miniaturization of SPADs, this greatly increases the system integration level and reduces the costs of the TOF sensors. A 128×128 single-photon TOF sensor with column-level time-to-digital converter array has been reported and achieved a depth image with high resolution . However, the SPAD-based TOF sensor still suffers from several design challenges. First, the dark noise, which is characterized by Dark Count Rate (DCR), limits the dynamic range of the sensors . Second, the SPAD needs to be quenched and recharged after avalanche. A general way to employ a SPAD is through a Passive-Quenching Circuit (PQC) , however the long dead time of PQC limits the maximum admissible photon counting rate and does not allow adjustable dead time to reduce afterpulsing . Passive quench with active reset circuit to reduce afterpulsing has been report  at the expense of area which not conducive for high fill factor integration.
In this paper, we present a linear pixel TOF sensor, which includes 16-pixel SPADs and 16 13-bit TDCs. The structure of the SPAD is optimized to reduce the dark noise by fabricating a poly guard ring to separate the Shallow Trench Isolation (STI) and the active area. An active quench and recharge circuit is proposed to reduce the dead time by a feedback delay chain and no obvious afterpulsing is observed. Two counters are utilized in the dual-counter based TDC to prevent the metastability of the counter in the TDC and thus avoid the counting errors occurred during the counters′ state transition.1 System architecture
Fig. 1 shows the block diagram of the TOF sensor chip. It consists of 16 SPADs, 16 quench and recharge circuits, 16 TDCs and the readout circuit. The linear array of SPADs operating in Geiger mode sense the photons reflected from objects, the quench and recharge circuits halt the SPAD's avalanche effect and recover the device to Geiger mode. The SPAD pixels output a train of digital pulses, which are fed into TDCs as the STOP signals. The START signal is synchronized with the emitted laser pulse. TDCs calculate the time difference between the START and STOP to achieve the time-of-flight. The TDC outputs are decoded to a 13-bit signal by the decoders and read out in parallel through the readout circuit.
SPAD is an avalanche diode that operates above the breakdown voltage, which is called as the Geiger mode. A good performance SPAD should has a low dark noise and high photon detection efficiency, which requires a high electronic field in the multiplication region. When operating in Geiger mode, particular design is required to prevent breakdown at the edge of the photon sensitive area. The SPAD cross section of this paper is illustrated in Fig. 2.
We prefer an octagon SPAD structure to eliminate junction edges and the diameter of the active area is 10 μm. The P+/N-well junction is fabricated in deep N-well to isolate substrate noise, with a P-well guard ring bordering the active area to prevent the Premature Edge Breakdown (PEB). Meanwhile, due to the STI can cause interface traps and increase the dark noise, a ring of thin-oxide poly around the edge of the guard-ring is formed to separate the STI and the active area. For continuous photon detection, the SPAD requires to be quenched and recharged. Quenching is able to halt the SPAD's avalanche breakdown process. After quenching, SPAD is also needed to be recharged and returns to Geiger mode again for next photon detection. The photons cannot be detected by the SPAD during the period of time before the device recovery, this duration, is called as dead time. Obviously, in order to improve photon counting rate, the dead time needs to be as short as possible, but it will make the device prone to afterpulsing which introduces coherent noises. To find a balance between high photon counting rate and low afterpulsing, the quench and recharge circuit with adjustable dead time is necessary. In this paper, we designed an active quench and recharge circuit with a feedback loop containing a delay unit to regulate dead time.
The transistor-level schematic of the active quench and recharge circuit is shown in Fig. 3(a). The P+ anode of each SPAD is connected to the quench and recharge circuit.VQCH and VB are generated on chip to supply bias voltages for the current sources. The cathode of the SPAD is biased higher than the SPAD′s breakdown voltage (VBD), in this state, an excess bias (VE) above VBD is maintained on the device, so that all the carriers in the multiplication region are depleted. Since the anode of the device holds on a high impedance state, when the SPAD is triggered, the avalanche current flows into the discharge current branch which is formed by M1, and the voltage of VP starts to raise rapidly, the bias voltage of the SPAD is hence reduced and the device is quenched. Simultaneously, the high level voltage of VP is fed back to open the large discharge branch after a few nanosecond delay and to speed up the recharge of SPAD, allowing the device to recover to the Geiger mode more quickly, the delay chain biased by VB controls the opening duration time of the large discharge branch. Four current branches controlled by EN [3:0] is designed to adjust the large current source value so that an adjustable dead time can be achieved, as can be seen in Fig. 3(b). An auxiliary path by M1 is designed with smaller current conducting capability. This branch adjusted by VQCH is helped to recharge the SPAD if the main discharge branch is completely shut down. The trigger pulse on VP during quenching is finally transformed to a digital signal through a digital buffer. The rising edge of this pulse accurately indicates the photon arrival time.
The 16 TDCs calculate the time difference between the START and STOP signals. The START signal is synchronized to the laser pulse and the STOP signals are the SPAD pulses. The 16 TDC output signals read out in parallel through the readout circuit as the TOF signals. The TDC structure is described as follows.
The architecture of the time-to-digital converter is show in Fig. 4. Four pairs of fully differential signals are generated by PLL and globally supplied to the TDC array. The on-chip PLL is composed by a generally used Phase-Frequency Detector (PFD), Charge Pump (CP) and current-controlled oscillator which is shown in Fig. 4(a). The reference clock of 75 MHz is fed by an off-chip crystal oscillator. Based on the 8-stage ring current-controlled oscillator, PLL produces 8-phase signals φ [7:0]. The operating frequency of PLL is 300 MHz, and the time delay between every two adjacent phase signal is 416 ps. φ [7:0] are buffer and enhanced by "CLKBUF" as indicated in the figure and distributed to all the 16-TDCs.
Operation of TDC is described as follows. The START and STOP interpolators sample PHI [7:0] at the time when START and STOP signal are generated. Two thermos codes of the interpolators are converted into 3-bit binary codes INT (START) [2:0] and INT (STOP) [2:0] respectively, which ultimately input to a TDC decoder as the six least significant bits of TDC_DIN, i.e., TDC_DIN [2:0] and TDC_DIN [5:3]. TDC_DIN [2:0] and TDC_DIN [5:3] represent the time residue of the input within one clock period and the counters calculate the cycles of the clock between START and STOP.
In order to avoid the metastable state of counters, two 10-bit counters are respectively counted by clock PHI  and PHI . When the sampling of STOP occurs during the state transition of either counter, the sampling errors of counter_0 or counter_1 maybe occurred. Since the count clocks of the two counters are opposite in phase, when the state of one counter is in transition, the other counter state must be safely stable. As a result, the output of counter_0 or counter_1 is assigned to the TDC_DIN safely depending on the phase of INT (STOP) [2:0]. The output of these counters are input to the TDC decoder as the ten most significant bits of TDC_DIN, i.e., TDC_DIN [15:6]. The 16-bits signal is finally computed by the TDC decoder into a 13-bit serial output which indicates the time difference between the START and STOP.
The counter timing diagram of TDC is shown in Fig. 4(c), as an example, the state of START and STOP are 00011110 and 11110000 respectively, after converting to binary code, INT(STOP)=3′d7, INT(START)=3′d2, according to the decode logic shown in Table 1, TDC_OUT=13′d5.
The chip has been implemented in 180 nm standard CMOS technology. The microscope photograph of the chip is shown in Fig. 5. The area of the chip is 3 mm×1 mm. The performance of the chip was tested and evaluated as follows.
The sensitivity of the device is characterized by Photon Detection Efficiency (PDE). To get the PDE of the SPAD, a commercial standard single photon diode with known photon detection efficiency is used as a reference. The reference device and the SPAD under test are placed separately besides the two light-outputs of a separately sphere which is illuminated by a wavelength tunable light source. The PDE of the tested SPAD is calculated by comparing the output counts to the standard device (except the dark counts) in uniformly illumination at different wavelengths. The plot of the PDE at 1 V excess bias voltage is shown in Fig. 6(a). The highest PDE of the SPAD is 18% at 550 nm wavelength.
The noise performance of SPADs is characterized by Dark Count Rate (DCR), which is due to thermally and tunneling generated carriers in P-N junction. Dark counts and photon triggered pulses cannot be directly distinguished, thus limiting the dynamic range of the sensors. Dark count rate of each pixels will be different due to the process variation. In this paper, the DCR of 16 pixels are measured for better evaluation of noise characteristics. The measured DCR distribution over all the 16 pixels, at different excess bias voltage is shown in Fig. 6(b). At room temperature, the median value of DCR is 8 kHz at 1 V excess bias voltage.
To verify the STI influences on the DCR, a device without poly isolation is fabricated to compare with the device depicted in Fig. 1 as can be seen in Fig. 6(c), the SPAD with poly guard ring shows lower DCR which proves that the poly guard ring can effectively reduce the dark noise.
The function of the active quench and recharge has been verified and the dead time measurement is shown in Fig. 6(d).The output waveform of the SPAD after quench and recharge is measured by an oscilloscope.When EN=0000, the feedback loop shut down and the SPAD current mainly discharges through the small discharge branch, the measured dead time is 30 ns; when EN=1111, the high level voltage of VP is fed back to open the large discharge branch after a few nanosecond delay thus to speed up the recharge of SPAD, as a result, the dead time can be reduced to as short as 8 ns and no obvious afterpulsing is observed.3.2 TDC measurement
In order to test the performance of the TDC, the measurement is under the external trigger mode without the SPADs. The resolution of TDC is 416 ps when the PLL is operating at 300 MHz. The START and STOP signals are provided by a digital delay generator which can generate a set of delay signals with picosecond resolution. TDC uses its nine least significant bits in actual ranging measurements, the measurement result is shown in Fig. 7(a). The test result shows an excellent linearity and accuracy. There is no counting error occurred thus proves the dual counter structure can effectively solve the counting error of TDC.
The nonlinear characteristic of TDC is measured based on the uniform distribution of SPAD trigger pulses under uncorrelated light . The DNL and INL are calculated by histogram statistics of TDC code in nine bits. The TDC non-linearity measurement results show in Fig. 7(b) and (c). The DNL is +0.64 LSB to -0.42 LSB and the INL is +4.60 LSB to -0.23 LSB.3.3 Time-of-flight results
A complete imaging system with electrical system, optical elements, mechanical scanning system and software to be used in LiDAR applications is developed. The SPAD sensor is read out by a FPGA-development board which communicating with the chip at 50 MHz and store the TDC outputs on a double data rate SDRAM memory. A USB controller communicates with PC at a speed of 48 Mbit/s. The sensor chip with 70 pins is bonded on the PCB, which is in turn connected to the rest of the supporting circuits.
The ranging measurements are performed with a 640-nm pulsed laser operating at 100 kHz repetition rate, the average power of the laser is 4.5 μW and the pulse width at FWHM is 90 ps. A signal generator generates a set of synchronization signals to drive the laser and provide the START signal for the TDCs, respectively. The pulsed laser being collimated with a divergence angle of 1.2° is placed on the side of the sensor and illuminated onto the object at a specific angle. The unambiguous range can be measured is 1 m which limited by the laser power. Due to ambient light and the effects of dark noise, it′s hard to get the correct distance information in each time process, especially in the case of low laser power, so that Time-Correlated Single Photon Counting (TCSPC) is required. The target is measured up to 1 m, and each point is measured with 20 480 photons counts to obtain the accurate time-of-flight though histogram statistics. The TCSPC measurements are repeated ten times at each point, the linearity and precision of the system are characterized and shown in Fig. 8, revealing a maximum non-linearity is 1.9%, and worst-case precision is 3.8%, over the entire range. The non-linearity error of the ranging measurement is mainly due to the TDC resolution and non-linearity.
For depth imaging, a non-coaxial two-dimensional scanning imaging system is set up, as can be seen in Fig. 9(a). A telescope consisting of two convex lenses with different focal lengths is used to collect the reflection light and converge to the sensor chip. The object is placed at a distance of 0.5 m on a two-axis electronic stage controlled by PC so that a complete scanning image can be obtained. Based on the TCSPC algorithm, each pixel is measured with 20 480 counts and statistically obtained the abscissa of the histogram peak as the exact depth value. This scanning imaging system successfully implements centimeter depth accuracy images.
As can be seen in Fig. 9(b), a 320×160 depth imaging is acquired at 0.5 m distance, the two toy cats are placed with a distance of 7 cm. Due to the limited laser power, the image is obtained in dark environment to improve the signal to noise ratio. The depth of the surface cannot be distinguished limited by the resolution of TDC.
Table 2 summarizes the performance of the sensor in comparison with other SPAD-based LiDAR systems[13-15]. For pixel performance, this design achieves a lowest dead time thanks to the active quench and recharge circuit. Although the poly guard ring effectively reduced the dark noise, the dark noise may still be high due to the relatively poor doping quality of the standard process, dedicated CMOS process may be considered in the future design. For LiDAR performance, our design achieves a higher spatial resolution of 320×160, and centimeter-accurate depth resolution at very low signal illumination. Meanwhile the sensor has the lowest power consumption. The TDC is suitable for remote sensing but not fit for the surface feature imaging, the structure may be optimized to achieve higher resolution for more applications.
A 16×1 pixels SPAD-based direct TOF sensor was implemented in 180 nm CMOS standard technology. System circuits include 16-pixel SPADs with active quench and recharge, and a dual-counter-based TDC is coupled with each SPAD.The experimental results show that the 10 μm diameter SPAD has a maximum photon detection probability of 18% at 550 nm in 1V excess bias voltage. Device noise performance is reported via a measurement of dark count rate distribution over all the pixels. The median dark count rate across all the devices in a sensor is 8 kHz at room temperature. The TDC is counted by 8 phase clocks based on a PLL operating in 300 MHz. 416 ps accuracy and excellent linearity are acquired. The DNL is +0.64 LSB to -0.42 LSB and the INL is +4.60 LSB to -0.23 LSB. For ranging and depth imaging, a 640 nm laser source is used to illuminate the scene up to 1 m with an average power of 4.5 μW. Accurate distance measurements are repeatedly achieved based on the TCSPC. The maximum nonlinearity error in distance measurement is 1.9% and the worst-case precision is 3.8% over the full measurement range. The sensor is enabled to reconstruct a 320×160 depth image with centimeter precision in low signal exposure. To get better performance, dedicated CMOS process for low SPAD dark noise may be considered, and TDC with better resolution and precision may be effectively improved the ranging precision and linearity.
NICLASS C, SOGA M, MATSUBARAH, et al. A 100-m range 10-frame/s 340×96-pixel time-of-flight depth sensor in 0.18μm CMOS[J]. IEEE Journal of Solid-State Circuits, 2013, 48(2): 559-572. DOI:10.1109/JSSC.2012.2227607
COVA S, LONGONI A, ANDREONI A. Towards picosecond resolution with single photon avalanche diodes[J]. Review of Scientific Instruments, 1981, 52(3): 408-412. DOI:10.1063/1.1136594
HORAUD R, HANSARD M, EVANGELIDIS G, et al. An overview of depth cameras and range scanners based on time-of-flight technologies[J]. Machine Vision and Applications, 2016, 27(7): 1005-1020. DOI:10.1007/s00138-016-0784-4
NICLASS C, FAVI C, KLUTER T, et al. A 128×128 single-photon image sensor with column-level 10-bit time-to-digital converter array[J]. IEEE Journal of Solid-State Circuits, 2008, 43(12): 2977-2989. DOI:10.1109/JSSC.2008.2006445
GUERRIERI F, TISA S, TOSI A, et al. Two-dimensional SPAD imaging camera for photon counting[J]. IEEE Photonics Journal, 2010, 2(5): 759-774. DOI:10.1109/JPHOT.2010.2066554
TISA S, GUERRIERI F, ZAPPAF. Variable-load quenching circuit for single-photon avalanche diodes[J]. Optics Express, 2008, 16(3): 2232-2244. DOI:10.1364/OE.16.002232
COVA S, GHIONI M, LACAITA A, et al. Avalanche photodiodes and quenching circuits for single-photon detection[J]. Applied Optics, 1996, 35(12): 1956-1976. DOI:10.1364/AO.35.001956
LIU M, HU C, CAMPBELL J C, et al. Reduce afterpulsing of single photon avalanche diodes using passive quenching with active reset[J]. IEEE Journal of Quantum Electronics, 2008, 44(5): 430-434. DOI:10.1109/JQE.2007.916688
MARKOVIC B, TISA S, VILLA F A, et al. A high-linearity, 17ps precision time-to-digital converter based on a single-stage vernier delay loop fine interpolation[J]. IEEE Transactions on Circuits and Systems Ⅰ:Regular Papers, 2013, 60(3): 557-569. DOI:10.1109/TCSI.2012.2215737
ZHANG C, LINDNER S, ANTOLOVIC I M, et al. A 30-frames/s, 252×144 SPAD flash LiDAR with 1728 dual-clock 48.8-ps TDCs, and pixel-wise integrated histogramming[J]. IEEE Journal of Solid-State Circuits, 2018(99): 1-15.
BULLER G, WALLACE A. Ranging and three-dimensional imaging using time-correlated single-photon counting and point-by-point acquisition[J]. IEEE Journal on Selected Topics in Quantum Electronics, 2007, 13(4): 1006-1015. DOI:10.1109/JSTQE.2007.902850
KOSTAMOVAARA J, HUIKARI J, HALLMAN L, et al. On laser ranging based on high-speed/energy laser diode pulses and single-photon detection techniques[J]. IEEE Photonics Journal, 2015, 7(2): 1-15.
VILLA F, LUSSANA R, BRONZI D, et al. CMOS imager with 1024 SPADs and TDCs for single-photon timing and 3-D time-of-flight[J]. IEEE Journal of Selected Topics in Quantum Electronics, 2014, 20(6): 364-373. DOI:10.1109/JSTQE.2014.2342197
NICLASS C, SOGA M, MATSUBARA H, et al. A 0.18μm CMOS SoC for a 100-m-range 10-frame/s 200×96-pixel time-of-flight depth sensor[J]. IEEE Journal of Solid-state Circuits, 2014, 49(1): 315-330. DOI:10.1109/JSSC.2013.2284352
PERENZONI M, PERENZONI D, STOPPA D. A 64×64-pixels digital silicon photomultiplier direct TOF sensor with 100-M photons/s/pixel background rejection and imaging/altimeter mode with 0.14% precision up to 6 km for spacecraft navigation and landing[J]. IEEE Journal of Solid-State Circuits, 2017, 52(1): 151-160. DOI:10.1109/JSSC.2016.2623635