1. Introduction
Recently, light detection and ranging (LiDAR) systems have been popular in various applications including augmented reality glasses, robot navigation systems, autonomous vehicles, indoor monitoring sensors for dementia patients [
1,
2,
3,
4,
5]. Typically, a pulsed time-of-flight (ToF) mechanism is exploited in LiDAR systems such that light signals with a pulse width of a few nanoseconds are transmitted to targets that are located within a feasible range. Then, the reflected pulses are detected and recovered by an optical receiver (Rx), and the distance to targets can be easily estimated by measuring the time interval between the transmitted START pulse and the received STOP pulse. Especially in an indoor-monitoring LiDAR sensor for senile dementia patients, where the detection range is limited to 5 m [
1].
In general, a linear-mode LiDAR sensor exploits avalanche photodiodes (APDs) whereas a Geiger-mode sensor employs single photon avalanche diodes (SPADs). Although the Geiger-mode SPADs provide high gain (~108), they typically mandate very high bias voltages, need a quenching circuit to remove after-pulses, and suffer from low photon-detection efficiency, high dark count rate, limited dynamic range, and timing jitters. Linear-mode LiDAR sensors using APDs can avoid these issues. Nevertheless, the linear-mode sensors yield much lower gain especially with on-chip CMOS APDs, thus limiting their usage to the short-range applications.
Figure 1 shows the block diagram of a ToF LiDAR system, in which a laser diode driver emits START pulses and the reflected light pulses are detected in the Rx which consists of an optical detector (i.e., avalanche photodiode or APD) to generate photocurrents, a transimpedance amplifier (TIA) to convert the photocurrents to voltage output signals, a post-amplifier (PA) to amplify the voltage outputs further, and finally, a time-to-digital converter (TDC) to compare the arrival time of the initial START pulse and the received STOP pulse, thereby estimating the distance to targets. This architecture, however, gives rise to inherent defects in the optical measurements. Firstly, it necessitates a costly laser driver to emit narrow optical pulses [
6,
7,
8,
9,
10,
11]. Secondly, targets are necessary for optical testing. It should be noted that various targets with different reflectivities affect the measured results of the LiDAR Rx ICs [
7,
8,
9]. Thirdly, when the transmitter (Tx) emits the optical pulses to targets, it must simultaneously send the electrical START pulses into the Rx. However, this electrical interconnection includes the parasitic components from I/O pads and bond wires between Tx and Rx, thereby causing undesired delays and signal distortions. Fourth, the electrical START pulses yield almost upright rising edges, whereas the rising edges of the STOP pulses show the finite and declined slopes. Hence, the inherent time difference, also known as a walk error, cannot be avoided in this measurement [
9].
Therefore, we propose a CMOS optoelectronic receiver IC (CORIC) for simple low-cost testing of a LiDAR Rx, which comprises an analog front-end (AFE) circuit, a high-speed digital input buffer, a variable delay cell, and a two-dimensional (2D) modified TDC. In particular, the AFE can successfully recover narrow pulses within the detection range up to 10 m [
10]. It consists of a dual-feedback folded-cascode differential TIA (DFD-TIA), an active single-to-differential (ASD) converter, a cross-coupled inverter-based post-amplifier (CI-PA), and a two-stage differential amplifier with negative impedance compensation (TDA-NIC). It should be noted that the fully differential configuration from the input stage in the DFD-TIA improves the common-mode rejection ratio (CMRR) and the power supply rejection ratio (PSRR) characteristics, which are critical factors for the signal integrity of the receiver even with the slight sacrifice of noise current spectral density.
The key contributions of this work can be summarized as follows:
A simple low-cost methodology is proposed to test LiDAR receiver ICs, discarding the need of costly complicated compensation algorithms.
By exploiting a high-speed buffer and variable delay cells right before the TDC, both START and STOP pulses can provide very similar pulse shapes, thus effectively avoiding the walk error issue during the measurements.
Time-interval between two pulses can be easily determined by varying the number of the delay cells.
This paper is organized as follows.
Section 2 describes each circuit of the CORIC and the simulation results.
Section 3 demonstrates the low-cost measurement results. Then, conclusions are made in
Section 4.
2. Circuit Description
Figure 2 depicts the block diagram of the proposed CMOS optoelectronic receiver IC (CORIC), in which the AFE circuit includes an on-chip APD as an optical detector to avoid the undesired signal distortion occurred from the notorious parasitic components of the bond wire inductance and the parasitic capacitance of I/O pads. Additionally, it exploits a voltage-mode DFD-TIA for optical-to-electrical conversion, a CI-PA for gain boosting, and a TDA-NIC with active feedback for DC offset cancellation such that the fully differential signals (V
op and V
on) can be generated. Thereafter, one of the differential signals passes through a high-speed digital input buffer (HS-DIB) that converts the differential voltage signals to a single-ended output (V
IB_O) with a full swing. Then, V
IB_O enters a variable delay line to mimic the time-interval (i.e., 1~15 ns in this work) between START and STOP pulses. Finally, the following TDC produces 4-bit digital codes which correspond to the designated time-intervals.
2.1. AFE Circuits
Figure 2 depicts the schematic diagrams of the crucial AFE circuits (i.e., DFD-TIA, CI-PA, and TDA-NIC with offset cancellation). Previously, various AFE optical receivers have been suggested [
6,
7,
8,
9,
10,
11,
12,
13] where off-chip APDs were mostly integrated on PC-boards via bond-wires. As previously mentioned, this bond–wire interconnection may lead to the considerable increase of packaging costs in the cases of multi-channel receiver arrays, and mandate on-chip electrostatic discharge protection diodes (ESD) which, however, deteriorate the receiver bandwidth and worsen the noise performance. Therefore, on-chip APDs can be an efficient alternative to replace the off-chip APDs to avoid these issues. Nonetheless, silicon on-chip APDs respond to lights only operating at infrared wavelengths. Besides, the practical CMOS implementation of on-chip APDs results in low responsivity [
14]. Despite these shortcomings, we have optimized the AFE circuits with the exploitation of CMOS on-chip avalanche multiplication [
10,
15].
The crucial specifications of the AFE circuits include 85-dBΩ transimpedance gain, 700-MHz bandwidth, 15-pA/
noise current spectral density, and a 40-dB power supply rejection ratio for the proper operations of the following digital circuitry [
10].
2.2. High-Speed Digital Input Buffer
Thereafter, a high-speed digital input buffer (HS-DIB) is incorporated to convert the analog voltage signals from the preceding AFE circuit to full-swing digital signals so as to facilitate the timing estimations at the following TDC.
Figure 3 shows the schematic diagram of the HS-DIB, in which an NMOS self-biased differential amplifier is connected in parallel with a PMOS self-biased differential amplifier.
If V
op is higher than V
on, M
1 and M
9 are turned on while M
2 and M
8 are turned off. Then, the drain of M
4 is low and V
IB_O becomes high. If V
on is higher than V
op, M
2 & M
8 are on while M
1 & M
9 are off. Then, the drain of M
4 is high and V
IB_O becomes low. When V
IB_O is low, M
11 helps to pull the drain of M
4 to become completely high. Therefore, if V
op is higher than V
on, V
IB_O generates a rail-to-rail positive voltage signal, and vice versa. It should be noted that hysteresis is introduced by M
11 because the HS-DIB may detect and boost even noises as weak signals. Hence, when the output (V
IB_O) is low, M
11 is turned on and it helps the output of the differential amplifier go towards ground rapidly [
16].
2.3. Variable Delay Line
It is well known that LiDAR systems suffer walk errors significantly because an inherent difference exists in the slopes of rising edges between the START and STOP pulses. These walk errors lead to considerable timing errors, mandating a complex compensation algorithm. The START pulses are generated from the laser driver in a Tx while the STOP pulses enter the TDC via a number of paths including the AFE circuits in a Rx. Therefore, the slope of the STOP pulses becomes much lower than that of the START pulses. With this innate defect, it is very difficult to accurately measure the performance of newly proposed receivers unless the complicated compensation circuits are equipped. In this paper, we suggest a variable delay line which can eliminate the need of a costly laser diode driver in Tx and a walk error compensation circuit in Rx. This proposed variable delay line enables the measurement of the performance of Rx accurately with no risk of walk errors.
Figure 4a depicts the schematic diagram of the variable delay line, which consists of a series of inverter chains with MOS capacitors attached at the output of each inverter for the purpose of delay-control. The PMOS (M
12) at the output helps to sharpen the falling edges of the STOP pulses. The MOS capacitors are turned on and off by using a 4-bit thermometer-to-binary decoder circuit.
Figure 4b compares the output pulses (V
IB_O) of HS-DIB with the simulated output pulses of the variable delay line in the case of 10-ns delay, where it can be clearly seen that the slope of the rising edges in both pulses is quite similar. The required specifications of the delay line are the variable delays from 1 ns to 10 ns so that the short-range detection can be successfully conducted.
2.4. Modified 2D Vernier TDC
In this work, the modified 2D Vernier architecture is exploited as a TDC circuit, in which resettable T-latch circuits are utilized to convert narrow input pulses to wide digital signals [
17]. It can, therefore, avoid the notorious ambiguity of pulse overlapping between START and STOP pulses.
The specifications of the 2D Vernier TDC include the detection capability of the time intervals of 1–10 ns.
Figure 5 illustrates the block diagram of the modified 2D Vernier TDC, where two resettable T-latch circuits are employed to maintain the high-state pulses until the arrival of the next input pulse and thus, the output ambiguity of the pulse overlapping between START and STOP pulses can be significantly reduced. In addition, with the resettable T-latch circuits, there is no need for dual threshold comparators to measure the time intervals, hence helping further to omit the walk errors. Consequently, errors would only occur due to the delay offset in a delay cell. In this work, each delay cell was designed to provide the delays of 3 ns (=t
1) and 2 ns (=t
2), respectively, so that the timing resolution of the modified 2D Vernier TDC becomes 1 ns (=t
1 − t
2).
2.5. Post-Layout Simulation Results
Figure 6 depicts the simulated results of the proposed CORIC, showing the effective correction of the error bits and allowing for the time difference (ΔT) of 10 ns that corresponds to the detection range of 1.5 m. This example generates the 15-bit output data of 000001111111111, which is finally converted to a 4-bit binary code of 1010 via a thermometer-to-binary (T2B) encoder. As long as the design space of the delay cells is large enough, the modified 2D Vernier TDC is expected to detect the reflected pulses within longer ranges.
3. Chip Fabrication and Measured Results
Figure 7 shows the chip photo of the proposed CORIC and its test setup, where the chip core occupies the area of 841 × 268 µm
2. An 850-nm laser source driver (Seed LDD, Notice Korea Ltd., Anyang, Korea) with a laser diode (Qphotonics, Ann Arbor, MI, USA) was utilized to generate light pulses. DC measurements revealed 20.9 mW power dissipation from a single 1.8-V supply.
Figure 8 demonstrates the measured output pulses of the variable delay lines and finally, the measured 4-bit thermometer binary codes of the TDC output when the delay is set from 1 ns to 11 ns. It can clearly be seen that the STOP pulses of the variable delay line have almost identical shapes to the START pulses. For examples, the final 4-bit thermometer binary data of the TDC outputs result in 0100 which corresponds to the 15-bit 000001111111111 (i.e., Δt = 4 ns), whereas 1010 that corresponds to the 15-bit 000000000001111 (i.e., Δt = 10 ns). Hence, the proposed CORIC chip yields very similar START and STOP signals so that it can eliminate the notorious walk errors during the measurements for the AFEs of LiDAR sensors, as anticipated.
Table 1 compares the performance of the proposed CORIC chip with the previously reported LiDAR receivers. Ref. [
6] depicts a frequency-compensated voltage-mode inverter TIA with a very low noise current spectral density that dissipates high-power and demands a large reverse bias for an off-chip APD for a high 50-A/W responsivity. Ref. [
7] implemented an AFE with a separate analog-to-digital converter/time-to-digital converter (ADC/TDC) chip in a 65-nm CMOS, achieving long detection range with low power consumption. Yet, it utilized an off-chip APD with a 160-V bias voltage for 25-A/W responsivity. Ref. [
9] suggested a differential-voltage-mode TIA that demonstrated a high transimpedance gain and a very large maximum detectable input current. However, AC coupling capacitors and bias resistors were necessary for an off-chip APD with 40-A/W responsivity. Additionally, a separate TDC chip was required for a wide input dynamic range. Ref. [
10] realized a fully differential AFE with an on-chip CMOS P
+/N-well APD that revealed the successful recovery of narrow 1-ns light pulses. Still, it demanded a costly laser diode for light emission and there was no TDC implemented above all. Ref. [
11] improved timing dynamic range and sensitivity characteristics with dual-mode control. The input currents were switched either to a charge-sensitive amplifier or to a gated active load-assisted TIA, depending upon the intensity of input currents. In particular, walk errors were substituted for small input currents by reducing the time-constant of the pulse shaper. Ref. [
13] obtained the minimum noise by utilizing the transimpedance-to-noise method. However, the input dynamic range was quite limited to 24.9 dB because of the high transimpedance gain with no gain control.