A 3D imaging concept based on pulsed time-of-flight focal plane imaging is presented which can be tailored flexibly in terms of performance parameters such as range, image update rate, field-of-view, 2D resolution, depth accuracy, etc. according to the needs of different applications. The transmitter is based on a laser diode operating in enhanced gain-switching mode with a simple MOS/CMOS-switch current driver and capable of producing short (~100ps FWHM) high energy (up to nJ) pulses at a high pulsing rate. The receiver consists of 2D SPAD and TDC arrays placed on the same die, but in separate arrays. Paraxial optics can be used to illuminate the target field-of-view with the receiver placed at the focal plane of the receiver lens. To validate the concept, a prototype system is presented with a bulk laser diode/MOS driver operating at a wavelength of 870nm with a pulsing rate of 100kHz as the transmitter and a single-chip 9x9 SPAD array with 10-channel TDC as the receiver. The possibility of using this method as a solid-state solution to the task of 3D imaging is discussed in the light of the results derived from this prototype.
© 2016 Optical Society of America
A 3D imager is a device that can measure the spatial coordinates of points on a surface within its field of view. Thus, contrary to a conventional camera, which measures only the intensity of the radiant excitation of points within a 2D image surface, a 3D imager can determine the distance of each of the surface points with respect to a reference (Fig. 1). 3D imagers have traditionally been used in areas such as mapping, automation, automotive, inspection, quality control and quality assurance. The development of an autonomous or “driverless” car is an attractive example of an application that obviously calls for high-speed environment-sensing techniques [1–3].
There has recently been a growing interest in using 3D techniques in a wider field of applications, including robotics, security and UAVs (Unmanned Aerial Vehicles). In addition, low-cost 3D imaging has great potential in consumer electronics (games), man-machine interfaces, surveillance and the control of machines (gesture control), for example . Many of these forth-coming applications require low cost, a high measurement speed and in many cases realization in miniature sizes quite comparable to modern CMOS cameras. One attractive solution is laser scanning, but the current commercially available laser scanners typically rely on a mechanical scanning concept, whereas the laser beam or several parallel beams can be scanned over the measurement volume mechanically. Although this technique achieves adequate performance for many applications, the systems based on it are typically relatively bulky, consume a lot of power and are also expensive. Thus, they cannot be regarded as a generic solution to the problem of 3D imaging. One potential step forward in developing new generations of 3D laser imagers is to replace the mechanical scanner with a solid-state imaging approach, i.e. a system which would be fully electronic and without any moving parts.
Several solid-state 3D laser imager concepts have already been proposed, see [4–13] and references therein, but their performance has so far been limited and they require quite sophisticated transmitters. We propose here a solid-state 3D laser imager architecture which might have potential as a generic solution in the sense that it is scalable in terms of performance parameters (range, accuracy and field-of-view) according to the particular needs of a specific application. The approach adopted here is based on the focal plane solid-state imaging technique, in which the round-trip transit time of a short, energetic laser pulse transmitted into the illumination cone of the system is measured with a 2D single photon detector array and a TDC (Time-to-Digital Converter) array. The detector array is located at the focal plane of a positive lens and can be realized with SPADs (Single Photon Avalanche Diodes) in a standard CMOS technology. The 3D imager concept based on a sub-ns laser pulse and time-of-flight measurement of the single photon events with a SPAD array is well-known, e.g. from [10–13], although previous realizations have used laser transmitter techniques, which did not fit well with the miniaturization concept. The realization presented here uses a miniaturized semiconductor laser diode (LD)-based transmitter operating at a high repetition rate and, more importantly, producing a short and yet energetic laser pulse (~1nJ/~100ps). This makes a single shot precision of ~100ps (<2cm in distance) available within a wide measurement range. The use of a semiconductor LD as the transmitter also paves the way for radical miniaturization of the whole 3D imager. Another important advantage of this approach is its digital nature (“impulses” sent, photons detected), which enables high measurement speed and accuracy with a relatively simple and robust realization.
This paper will first explain the architecture of the proposed solid-state 3D imager in detail. It will then present the structure of a low resolution prototype imager and quote some measurement results as a validation of the concept. Finally, the achievable goals and advances associated with this concept will be discussed.
2. The solid-state 3D laser imager concept
2.1 General architecture
The solid-state 3D laser imager approach based on pulsed TOF (Time-of-Flight) laser ranging is presented in Fig. 2. An impulse-like burst of photons from a LD transmitter illuminates a target area within the FOV of the system. A portion of the returning photons are detected by a 2D receiver array located at the focal plane of a positive lens (system), and the transit times (Δt) of the detected photons are measured. The 2D receiver array maps the target surface in an xy plane (perpendicular to the optical axis) and Δt is translated to distance into the direction z (the optical axis). The system can use paraxial optics of a modest size with matched divergence in the transmitter and receiver.
The 2D receiver uses SPADs realized in a standard CMOS technology together with TDCs on the same die. A SPAD is simply a diode reverse biased above its breakdown voltage. As a result, any stimulus leading to the creation of an electron-hole pair (e.g. photon absorption) within the planar multiplication region may induce avalanche breakdown, resulting in a “digital-like” output signal . A SPAD detector or detector array can be realized with varying active areas and configurations. CMOS SPADs have low timing jitter (50-100ps) and wavelength dependent PDP (Photon Detection Probability) typically of 2-30% (corresponding to 870nm-550nm) and do not require analog signal processing, contrary to that needed with a linear APD (Avalanche Photo-Diode) receiver [14–19].
In addition to the timing jitter of the detectors (σSPAD), there are two other main factors contributing to the overall timing (distance measurement) uncertainty of a pulsed TOF laser 3D imager: the time interval measurement uncertainty (σTDC) and the duration of the laser pulse (σLD). Since all these factors are independent when operating in single photon mode (single photon incidents causing avalanche events in detectors), the overall uncertainty can easily be estimated by √(σSPAD2 + σTDC2 + σLD2). Timing jitter in the detector is intrinsic to a CMOS SPAD and is 50-100ps depending on the CMOS technology and SPAD structure used [14–19]. A timing uncertainty of the same order in other sources of jitter can be considered an optimal compromise between performance and cost/complexity, implying a laser pulse FWHM of ~100ps (σLD = ~42.5ps) and a TDC precision of ~50ps. This would correspond to centimeter level single shot precision in terms of distance, bearing in mind that further averaging of successive measurements would improve the precision by a factor of √N when considering N measurements.
On the other hand, in order to be able to measure at high speed, e.g. 10-25 image updates per second, and to have simultaneously a reasonable measurement FOV and range (e.g. 25ᵒx45ᵒ in 5-50m), the energy of the laser pulse should be relatively high. In principle, gain switching can be used in semiconductor LDs to produce high-speed pulses with a width of <100ps [20–22], but unfortunately, the peak pulse power with standard LDs in the gain-switching regime is limited to a few hundreds of mWs, and even this can only be achieved with quite sophisticated ultra-fast laser driver techniques. A straightforward compensation of the low peak power with a higher pulsing rate requires higher time interval measurement rate capability. And more importantly, it increases the number of unwanted detections by the SPADs, leading to a longer measurement time. A SPAD can be randomly triggered due to thermal or tunneling-generated electron-hole pairs and also due to possible background illumination. This is in fact one of the major issues in single photon detection-based laser ranging . Due to random triggering of the SPAD detector, a histogram analysis of the triggerings is needed, implying that valid detection of the target usually requires detection of more than one laser shot. With high background illumination, the target may even be blocked due to random photons occurring between the emitted laser shot and the photons backscattered from the target. One possibility to circumvent blocking is to use time-gating of the SPAD detector .
2.2 The prototype 3D imager
A graphical representation of the prototype developed here is shown in Fig. 3(a) and a photograph of the physical implementation in Fig. 3(b). The measurement system uses a fiber and a collimator together with an engineered diffuser on the transmitter side to achieve homogeneous illumination of a square with a divergence of 24.5mrad and a positive lens on the receiver side with an effective diameter of 18mm and focal length of 20mm. Thus, the divergence of the receiver is approximately 15mrad. A narrow-band interference optical filter (NBOF) is used to limit the optical bandwidth of the receiver to ~40nm in order to accommodate the spectral drift of the laser pulse and block most of the background at the same time. An FPGA interface is used to transfer the measurement data to a PC, where MATLAB is used for data analysis and image construction. The FPGA also works as a control unit, providing a triggering signal for the LD driver and configuration data for the receiver IC.
2.2.1 The laser diode-based transmitter
As discussed above, the critical system level parameters of the laser transmitter are the energy level and width of the optical pulse, the pulsing rate and the overall complexity of the system. A high energy (~nJ) pulse is required to realize a measurement distance of a few tens of meters to non-cooperative targets (Lambertian) with a reasonable measurement time and optics size (receiver lens diameter <2-3cm) . A short pulse width of 100-150ps (FWHM) is required to realize a high single shot distance measurement precision. A change of 1cm in the position of the target under measurement corresponds to approximately 67ps in the measured transit time of the photons, and thus the above pulse width corresponds roughly to a distance measurement precision of ~2cm in a single measurement (single shot). A high pulsing rate is advantageous for a short measurement time, especially in a single photon detection mode, because the time needed for collecting the required number of detections from the target is shorter considering a high pulsing rate. On the other hand, an unambiguous measurement requires that a new laser pulse is not sent before the echo of the previous pulse is received. Thus, for a distance range of 100m, for example, the maximum allowed rate is ~1.5MHz without a more complex modulation scheme .
In order to have a less complex system, a semiconductor LD-based transmitter would be preferred. Gain switching is a previously known method for producing high-speed impulse-like pulses from a semiconductor laser diodes and is based on a concept in which the internal resonator of the LD is brought into relaxation oscillations with a high-speed drive current edge. If the width of the drive current pulse is less than the distance of the successive pulses of the relaxation oscillations, a single isolated pulse can in principle be achieved [20–22]. However, the pulse energy is typically only in the ~10pJ range in the standard gain switching mode .
Recently, it was shown that with a semiconductor LD with a large equivalent spot size (thickness of the active region/confinement factor, da/Γa>>1), gain switching can be enhanced resulting in a higher optical power and energy [26,27]. It was also shown that this modification can be applied to a DH (double heterostructure) laser diode structure e.g. by shifting the position of the active layer from the peak of the asymmetric transverse optical mode [26–29]. The large da/Γa allows for a large number of carriers (and thus higher energy) to be accumulated before the onset of the optical pulse, while positioning the high-energy optical pulse at the trailing edge of the pumping pulse.
The present laser diode transmitter consists of a specially designed bulk GaAs/GaAlAs DH (double heterostructure) laser diode operating at a wavelength of ~870nm and a MOS current driver. The internal LD construction follows the above principles and has a large equivalent spot size (d/Γ = ~4um) in order to enhance gain switching, with the result that a single high-energy (~1nJ) and high-speed (~125ps FWHM) optical output pulse is produced from the laser diode with active dimensions of 30µm (width of the active stripe) and 3mm (length of the optical cavity). This power/energy is substantially higher than the levels available from standard LDs with similar dimensions in the standard gain-switching mode. Moreover, the drive current parameters needed to produce the above mentioned optical pulse are 6.5A and ~1ns for the peak current and current pulse width, respectively. In other words, the drive current pulse width can be considerably longer than the optical pulse width simplifying the design of the driver electronics. Here, the driving current of the LD was produced with a MOS-based driver and an LCR transient-based pulse shape control as is explained in greater details in [30,31]. The pulsing rate of the transmitter is 100kHz in the measurement setup, but can be increased up to 1MHz. In order to provide a low-jitter electrical start timing signal for time interval measurements, the current of the laser diode is sampled to eliminate the jitter of the LD driver as opposed to the FPGA providing the start signal. The LD chip is wire-bonded on a ceramic substrate, which is then mounted with a conductive glue onto the printed circuit board. The whole LD transmitter electronics occupies a circuit board area of 2-3cm2 only, as shown in Fig. 3.
2.2.2 The receiver
The detector/measurement IC detects the photons reflected from the target and measures their flight times. The receiver chip (see photograph in Fig. 4) consists of a 9x9 SPAD array together with a 10-channel TDC, realized in a 0.35μm HV CMOS technology. The size of the detector array is 330μm x 330μm with each detector’s active area of size 24μm x 24μm, corresponding to a fill factor of ~43%. Each SPAD has a square active area with rounded corners and a deep-nwell/p + junction as a multiplication region. Premature edge breakdown is avoided by using a pwell guard ring around the active area. To achieve a higher fill factor, all 81 pixels share one common deep-nwell (detections are read from the anodes), and all the front-end electronics are placed outside the SPAD array.
The TDC has 10 parallel channels for time interval measurement, one for receiving the start signal and the remaining 9 for receiving timing signals from the SPAD matrix (stops). The TDC is realized so that any of the 49 possible 3x3 sub-arrays can be connected to the 9 TDC channels, and the selection can be changed while measurement is in progress. Thus, 9 scans of 3x3 are needed in order to measure the whole 9x9 array. Time interval measurement is based on a counter and delay line interpolation using two nested levels with different resolutions. The measurement architecture is presented in Fig. 5.
The 20MHz input reference signal is multiplied internally to 240MHz with a multiplying delay-locked loop (MDLL) [32,33] consisting of an adjustable delay-line, a phase detector and a charge pump. The phase detector monitors the delay in the delay-line and the charge pump adjusts it so that the cycle time of the delay-line (multiplied by 12) corresponds to the cycle time of the reference crystal. This delay locking prevents measurement fluctuations caused by process, voltage and temperature (PVT) variations. The start signal sets the 7-bit counter and the 9 stop signals take samples of the state of the counter. Hence, the counter solves the intervals between the timing signals with a resolution of 4.2ns. The maximum measurement range is 530ns, which corresponds to ~80m in a pulsed laser distance measurement.
The MDLL delay line interpolates the locations of the timing signals (start and 9 stops) within the counter cycle time. A delay-line with identical successive delay elements generates 16 even-sized phases with a resolution of 260ps when the multiplied reference signal propagates in the MDLL with a frequency of 240MHz. The timing signals store the prevailing state of the delay-line in a register bank, from where the first interpolation level results can be decoded.
The second interpolation level continues the interpolation and solves the locations of the timing signals within the first interpolation level delay element. One delay period is divided into 32 LSB phases with parallel load capacitor-scaled delay elements . The small time differences of 8.1ps between the parallel elements are achieved by means of a small difference in their capacitive loads, leading to mm-level TDC resolution. The TDC was intentionally overdesigned in order to measure the characteristics of the SPAD array, but, in order to be in line with the concept presented above, the first 3 LSBs of the TDC results are ignored here, bringing the resolution to ~65ps. The TDC has a power consumption of 150mW when operating at a clock frequency of 240MHz, the main source of power consumption being the continuously circulating MDLL.
The receiver IC is capable of operating in both free running and gated mode, the principle of the latter being shown in Fig. 6. At the beginning of each measurement cycle, one of the 49 possible 3x3 sub-arrays is selected for measurement and the position and width of the gate window is defined with a resolution equal to the internal clock period of the TDC with respect to the start signal (the laser pulse). After the arrival of the start signal, the 9 chosen SPADs are biased above breakdown for the predefined gate window in order to be ready for detecting incident photons, while all the others are kept quenched during the whole measurement cycle. Upon detection of a photon, one or more SPADs send stop signals to the TDC at the rising edge of the anode voltage. The TDC measures the transit time of the photons detected from all 9 SPADs simultaneously with respect to the start channel. If no triggering occurs due to photon detection or internal noise, the SPADs are actively quenched at the end of the gate window (Fig. 6, case 1). The measured data are transferred out of the chip, once the Ready signal is raised by the IC. The measurement data include 9 16-bit words, with each word defining the interval measurement of one channel. Each word is read out in two bytes (8bit data path), and the maximum data transfer rate is 66Mbytes/s. The gate window and sub-array selection settings can be changed after the transfer of the measurement data and before the arrival of the next Start rising edge. Sel InOut determines the direction of data transfer. A frame update rate of ~11kHz (100/9) can be reached with a laser pulse frequency of 100kHz.
Cases 2 and 3 in Fig. 6 demonstrate how gating can be used to avoid background triggerings from flooding of the SPADs. Gating has been shown to be an effective way of suppressing background noise even with high background illumination (in outdoor sun light) with target distances up to several tens of meters . Various algorithms can be developed for controlling the placement and width of the gate window depending on environment conditions such as target movement, background illumination, etc., in order to reach the highest possible image update frequency.
The single shot distribution of one SPAD operating in single photon mode with a flat, non-cooperative target at ~19m is presented in Fig. 7(a). The FWHM of the distribution is ~170ps, which matches well with the calculations in case of a 125ps laser pulse, 50-100ps SPAD jitter and ~30ps TDC uncertainty (dominated by the quantization noise of the TDC, ~LSB/√6). The distribution tail, which is mostly caused by the diffusion of carriers in the SPADs, accounts for nearly 30% of the detections. A detection is a triggering of a SPAD caused by a laser photon returning from the illuminated target, dark or background noise. In order to investigate the effect of averaging on precision, TOF was resolved by averaging N consecutive detections where N varied between 1 and 250. The measurements or each value of N were repeated 104 times and each time the TOF was derived by averaging. Then a distribution of the measured TOFs was formed for each value of N. The measurement conditions (temperature, background illumination, target distance and reflectivity, etc.) were kept stable for the whole duration of the measurements. The measured mean and sigma of the distributions versus the number of averaged detections (N) are shown in Fig. 7(b) for two cases that correspond to presence or filtration of the tail. The tail was filtered by assuming a symmetrical histogram with respect to the peak and removing the hits that fell outside the time period for this ideal histogram. Filtering the tail obviously reduces the value of the mean. In both cases the uncertainty improves by a factor of √N considering N measurements with respect to the single shot sigma. In order to achieve cm level accuracy, σ should be less than 67ps (corresponding to 1cm in distance measurement). Comparing the two cases, filtering the tail seems to improve the uncertainty by a factor of nearly 3, so that with the tail detections considered in averaging, around 350 detections of the target are needed, for example, to achieve the same result as 50 detections with omission of the tail (15 out of the 50 are omitted detections of the tail). These measurements were performed under laboratory conditions, so that the dominant source of noise was the dark count of the SPADs.
An average Dark Count Rate (DCR) of ~150kHz was measured throughout the array but there were a few SPADs having 10 times larger DCRs. Detailed measurement results for the receiver IC specifications can be found in .
Figures 8(a), 8(b) and 8(c) show the 3D images of three targets placed at a distance of ~19-20m, at which the FOV of the receiver covers roughly a 27x27cm square, implying that each pixel sees an area of approximately 3x3cm. The first target (Fig. 8(a)) is a white pyramid with three steps, the second (Fig. 8(b)) a white paper ramp with its surface at a 60-degree angle to the optical axis, one-third of which is covered with a lower reflective material (brown carton). The third target (Fig. 8(c)) is a flat plane of white paper with a cube placed on top, causing a ~3cm step. The distance for each pixel is resolved by considering 50 detections, filtering out the distribution tail hits, and averaging over the rest of the hits (~35). According to Fig. 7(b), this leads to a σ value of ~20ps (~3mm in distance). The measured distances are calibrated to compensate for the static delay difference between the detectors in the array (max delay ~20ps) caused by wiring delay mismatch in the receiver IC. Since this delay difference is constant and independent of measurement condition, the calibration is simply a subtraction of delay difference for each SPAD. The detectors were operating in the single photon mode. The imaging uncertainty is in agreement with the calculations and experimental results of Fig. 7 and is less than 3mm.
Despite having a lower measurement speed and, as a result, a lower image update rate, operation in the single photon mode has certain advantages. There is no walk error present and pile-up effects (which are present in the multi-photon detection mode [7, 24]) can be prevented. As a result, there is no need for walk error compensation schemes. Photon detection rate of a SPAD is inverse of the average time between photon detections and can be estimated using the radar equation . The detection rate of each SPAD in single photon mode simply gives a measure of the reflectivity of the spot measured on the target, because the effects of the distance of the target spot and the illumination pattern are known and can be accounted for. A color map representation of the photon detection rate for the ramp target after filtering of the dark count noise is shown in Fig. 8(d). The average measured detection probability of photons for white paper in this measurement was ~4.5% (corresponding to an image update rate of 10 images per second). The lower detection rates of the SPADs on the left side represents the lower reflectivity of left side of the target. Also, while operating in the single photon mode, the characteristics of the histogram of hits can be further analyzed to gather more information about the area seen by one pixel other than only a distance measure derived from averaging, e.g. the presence of a step within the field of view of the pixel.
The measurements presented above indicate the potential of this 3D laser imager concept as an adaptable low-cost solution for 3D imagers. The fiber/diffuser that was used in the current system as a simple way of producing an evenly distributed illumination can be replaced with a compact lens system. Also here, the divergence in the transmitter did not match well with that in the receiver, and as a consequence, almost more than half of the illumination power was lost (on top of the usual loss caused by the common inefficiency of optics). With more efficient optics and better matching of transmitter and receiver FOVs an optical efficiency of 70% could be expected.
LD operation in the enhanced gain-switching mode can enable the laser pulse energy to be increased to a few nJs, while the pulse width remains of the same order and the peak drive current is kept at a reasonable level (~10A). Also, the wavelength of the LD can be reduced to ~810nm, corresponding to ~5% SPAD PDP compared with only 2% when operating at 870nm. Some parts of the laser diode driver can be implemented in CMOS  to reduce the size of the laser imager even further. Thus, the 3D imager can in principle be quite small, perhaps about the size of a match-box.
On the receiver side, higher resolution 2D SPAD arrays can be implemented in CMOS up to kilo pixels, but fill factor gains in importance once such an array is considered. There are two main reasons for the loss in the fill factor: the DCR of the SPADs and a non-photosensitive area within the pixel pitch. The DCR increases super-linearly with any increase in the active area, and thus limits the maximum possible active area size. Also, in most SPAD array implementations with in-pixel electronics the trade-off between the performance of the pixel and the fill factor leads to an active area-to-pitch ratio as low as a few percent. Having the detector and TDC arrays on the same chip but placing the TDCs and front-end electronics outside the SPAD array is a simple and effective way of insuring a high fill factor. With a large array of a few kilo pixels in a 0.35µm CMOS technology the fill factor can be still quite high (30-40%) compared with many other architecture employing in-pixel TDCs or counting electronics, even when microlenses are used to increase the fill factor . Once a deep sub-micron CMOS technology (<0.25µm) is considered for implementing the receiver IC, implying an increased number of metal layers for routing and less limitations on the layout design rules, the active area of the SPAD can account for an even higher proportion of the pixel pitch.
Having hundreds or even thousands of TDCs on the same chip as a detector array can naturally result in a temperature increase due to high power consumption in the IC, and as a consequence, an increase in the temperature-dependent DCR, but, given an efficient gating mechanism, the DCR can be suppressed from blocking the detectors before the arrival of photons from the target, while TDCs can be made to operate only during the gate window, in order to reduce the power consumption. As power consumption increases with any increase in the frequency of operation, having a high energy laser pulse with a lower rate also eases the complexity and operation of receiver electronics, which will lead to a lower power consumption and thermal DCR. The DCR can also be reduced to hundreds of hertz with the new SPAD structures made possible by the retrograde nwell doping profile available in deep sub-micron CMOS technologies .
Extra jitter can be expected once a large array of thousands of elements is considered, due to digital switching and the longer signals and power routes on the chip, but since the SPADs are operating in single photon mode, a few percent of them are triggered simultaneously during one measurement, so that generated noise on substrate and VDD lines will not be a huge issue. Also, an optical start signal can be integrated into the system to eliminate any temperature-dependent timing drifts caused by the laser transmitter.
As far as the results are concerned, if the specifications of the system are improved as explained above, say a 4nJ laser pulse with a 100kHz pulsing rate, operating at 810nm (SPAD PDP of 5%) and a 4k SPAD array with a FF of 35% can be considered and the FOVs of the receiver and transmitter will be better matched to achieve an optical efficiency of 70%, for an imaging application with a 20m maximum range, whereupon the minimum detection rate per pixel (for a brown cardboard target with a reflectivity of ~50%) would be roughly 0.5%. If simple averaging is to be used for deriving the distance, 5 detections are needed to update each distance result in order to achieve an uncertainty of less than 1cm, implying that each pixel result will take 1000 laser shots to update. Thus, even if each TDC channel were shared between 16 SPADs, for example, an image update rate of 5-10 images per second could be achieved.
Although short gate windows (a few tens of nanoseconds) might be needed for background noise cancelation, quick scans can be used to perform the first sweeps with longer gate windows simply to locate the whereabouts of the target in the distance range, and then short gate windows can be used to resolve the exact position of the target with sub-cm-level uncertainty. Computational imaging approaches have also been developed recently that suggest methods for resolving the target distance with only a few detections (or even one) [36,37], which might be of help in increasing the image update rate.
As pointed out in the introduction, there are many former studies on the development of solid-state 3D imagers, see e.g [4–13]. and references therein. The anticipated performance depends strongly on the application details and thus, a general comparison of the various approaches is a demanding task. As mentioned above, in this work, the aim is to achieve a range of a few tens of meters with a spatial resolution of ~4k pixels, a distance measurement accuracy of ~1cm, and a frame rate of a few images per second, while putting emphasis on miniaturization of the 3D imager. With this performance, applications within the fields of robotics, automation and security should be reachable. The most important difference of this work with many prior 3D imagers is the transmitter construction. Whereas many approaches similarly “digitize” the receiver with CMOS SPAD techniques, the transmitter is in many cases still “analogue”. For example, the transmitter reported in  is based on an array of 15 continuous-wave (CW) modulated LDs with an average optical power of 1.5W. This, inevitably sets challenges to the miniaturization of the system and requires high power for the transmitter (~1.5A), although the presented results are otherwise promising. In [7, 10–12], pulsed laser sources with a peak optical power of 100-250mW at 635nm, a repetition rate of 40-50MHz and the pulse width of ~100ps (FWHM) is used to illuminate the target. The low pulse energy is partly compensated for by the high repetition rate, which adds to the complexity of receiving electronics and makes background noise a bigger issue. In our work, the goal is to “digitize” the transmitter as well by using a semiconductor laser diode. With the presented approach, i.e. using impulse-like laser pulses as the probe signal, the system complexity is markedly reduced since the inherent precision is defined by the short pulse width. At the pulsing rate of 100kHz, for example, the average optical power of the transmitter (with a ~1nJ/100ps pulse) is at the sub-mW level. At the same time, high pulse energy (nJ or more) is provided in order to achieve a range of tens of meters to non-cooperative targets within a reasonable measurement time.
Based on the results shown above, we believe that the recently suggested high energy/high-speed semiconductor LD concept based on the enhanced gain switching operation mode utilized in the direct pulsed TOF measurement with a CMOS SPAD/TDC array receiver may pave the way to the miniaturization of the future solid-state 3D imagers.
Academy of Finland (Centre of Excellence in Laser Scanning Research, contract no. 272196, and contract nos. 255359, 283075 and 251571); Finnish Funding Agency for Innovation (TEKES).
References and links
1. B. Schwarz, “LIDAR: mapping the world in 3D,” Nat. Photonics 4(7), 429–430 (2010). [CrossRef]
2. V. C. Coffey, “Imaging in 3-D: killer apps coming soon to a device near you!” Opt. Photonics News 25(6), 36–43 (2014). [CrossRef]
3. A. S. Malik, Depth Map and 3D Imaging Applications: Algorithms and Technologies: Algorithms and Technologies (IGI Global, 2011).
4. R. J. Walker, J. A. Richardson, and R. K. Henderson, “A 128×96 pixel event-driven phase-domain ΔΣ-based fully digital 3D camera in 0.13μm CMOS imaging technology,” in International Solid-State Circuits Conference Digest of Technical Papers (IEEE, 2011), pp. 410–412.
5. R. Lange, “3D time-of-flight distance measurement with custom solid-state image sensors in CMOS/CCD-technology,” Dissertation, Department of Electrical Engineering and Computer Science, University of Siegen (2000).
6. D. Bronzi, Y. Zou, F. Villa, S. Tisa, A. Tosi, and F. Zappa, “Automotive three-dimensional vision through a single-photon counting SPAD camera,” IEEE Trans. Intell. Transp. Syst. 17(3), 782–795 (2016). [CrossRef]
7. V. Mitev and A. Pollini, “Flash imaging sensors for space applications,” in 7th International Conference on Recent Advances in Space Technologies (IEEE, 2015), pp. 687–693.
8. O. Shcherbakova, L. Pancheri, G. F. Dalla Betta, N. Massari, and D. Stoppa, “3D camera based on linear-mode gain-modulated avalanche photodiodes,” in International Solid-State Circuits Conference Digest of Technical Papers (IEEE, 2013), pp. 490–491. [CrossRef]
9. M. Perenzoni, D. Perenzoni, and D. Stoppa, “A 64x64-pixel digital silicon photomultiplier direct ToF sensor with 100Mphotons/s/pixel background rejection and imaging/altimeter mode with 0.14% precision up to 6km for spacecraft navigation and landing,” in International Solid-State Circuits Conference (IEEE, 2016), pp. 118–119.
10. C. Niclass and E. Charbon, “A single photon detector array with 64× 64 resolution and millimetric depth accuracy for 3D imaging,” in International Digest of Technical Papers.Solid-State Circuits Conference (IEEE, 2005), pp. 364–604.
11. C. Niclass, A. Rochas, P. A. Besse, and E. Charbon, “Design and characterization of a CMOS 3-D image sensor based on single photon avalanche diodes,” IEEE J. Solid-State Circuits 40(9), 1847–1854 (2005). [CrossRef]
12. C. Niclass, C. Favi, T. Kluter, M. Gersbach, and E. Charbon, “A 128x128 single-photon imager with on-chip column-level 10b time-to-digital converter array capable of 97ps resolution,” in IEEE International Solid-State Circuits Conference-Digest of Technical Papers (IEEE, 2008), pp. 44–594.
13. G. Buller and A. Wallace, “Ranging and three-dimensional imaging using time-correlated single-photon counting and point-by-point acquisition,” IEEE J. Sel. Top. Quantum Electron. 13(4), 1006–1015 (2007). [CrossRef]
14. A. Rochas, A. R. Pauchard, P. A. Besse, D. Pantic, Z. Prijic, and R. S. Popovic, “Low-noise silicon avalanche photodiodes fabricated in conventional CMOS technologies,” IEEE Trans. Electron Dev. 49(3), 387–394 (2002). [CrossRef]
15. S. Burri, Y. Maruyama, X. Michalet, F. Regazzoni, C. Bruschini, and E. Charbon, “Architecture and applications of a high resolution gated SPAD image sensor,” Opt. Express 22(14), 17573–17589 (2014). [CrossRef] [PubMed]
16. D. Bronzi, F. Villa, S. Bellisai, B. Markovic, S. Tisa, A. Tosi, F. Zappa, S. Weyers, D. Durini, W. Brockherde, and U. Paschen, “Low-noise and large-area CMOS SPADs with timing response free from slow tails,” in Proceedings of the European Solid-State Device Research Conference (2012), pp. 230–233. [CrossRef]
17. J. A. Richardson, E. Webster, L. A. Grant, and R. K. Henderson, “Scalable single-photon avalanche diode structures in nanometer CMOS technology,” IEEE Trans. Electron Dev. 58(7), 2028–2035 (2011). [CrossRef]
19. F. Villa, R. Lussana, D. Bronzi, S. Tisa, A. Tosi, F. Zappa, A. Dalla Mora, D. Contini, D. Durini, S. Weyers, and W. Brockherde, “CMOS imager with 1024 SPADs and TDCs for single-photon timing and 3-D time-of-flight,” IEEE J. Sel. Top. Quantum Electron. 20(6), 364–373 (2014). [CrossRef]
20. K. Y. Lau, “Gain switching of semiconductor injection lasers,” Appl. Phys. Lett. 52(4), 257–259 (1988). [CrossRef]
21. D. Bimberg, K. Ketterer, E. H. Bottcher, and E. Scoll, “Gain modulation of unbiased semiconductor lasers: ultrashort pulse generation,” Int. J. Electron. 60(23), 23–45 (1986). [CrossRef]
22. S. Schwertfeger, A. Klehr, T. Hoffmann, A. Liero, H. Wenzel, and G. Erbert, “Generation of sub-100 ps pulses with a peak power of 65 W by gain switching, pulse shortening, and pulse amplification using a semiconductor-based master oscillator-power amplifier system,” Appl. Opt. 52(14), 3364–3367 (2013). [CrossRef] [PubMed]
23. S. Cova, M. Ghioni, A. Lotito, I. Rech, and F. Zappa, “Evolution and prospects for single-photon avalanche diodes and quenching circuits,” J. Mod. Opt. 51(9–10), 1267–1288 (2004). [CrossRef]
24. J. Kostamovaara, J. Huikari, L. Hallman, I. Nissinen, J. Nissinen, H. Rapakko, E. Avrutin, and B. Ryvkin, “On laser ranging based on high-speed/energy laser diode pulses and single-photon detection techniques,” IEEE Photonics J. 7(2), 1–15 (2015). [CrossRef]
25. N. J. Krichel, A. McCarthy, and G. S. Buller, “Resolving range ambiguity in a photon counting depth imager operating at kilometer distances,” Opt. Express 18(9), 9192–9206 (2010). [CrossRef] [PubMed]
26. B. Ryvkin, E. A. Avrutin, and J. T. Kostamovaara, “Asymmetric-waveguide laser diode for high-power optical pulse generation by gain switching,” J. Lightwave Technol. 27(12), 2125–2131 (2009). [CrossRef]
27. B. S. Ryvkin, E. A. Avrutin, and J. T. Kostamovaara, “Quantum well laser with an extremely large active layer width to optical confinement factor ratio for high-energy single picosecond pulse generation by gain switching,” Semicond. Sci. Technol. 26(4), 045010 (2011). [CrossRef]
28. B. S. Ryvkin, E. A. Avrutin, and J. T. Kostamovaara, “Vertical cavity surface emitting lasers with the active layer position detuned from standing wave antinode for picosecond pulse generation by gain switching,” J. Appl. Phys. 110(12), 123101 (2011). [CrossRef]
29. B. Lanz, B. S. Ryvkin, E. A. Avrutin, and J. T. Kostamovaara, “Performance improvement by a saturable absorber in gain-switched asymmetric-waveguide laser diodes,” Opt. Express 21(24), 29780–29791 (2013). [CrossRef] [PubMed]
30. L. Hallman, J. Huikari, and J. Kostamovaara, “A high-speed/power laser transmitter for single photon imaging applications,” in Proceedings of IEEE SENSORS (IEEE, 2014), pp. 1157–1160.
31. J. M. T. Huikari, E. A. Avrutin, B. S. Ryvkin, J. J. Nissinen, and J. T. Kostamovaara, “High-energy picosecond pulse generation by gain switching in asymmetric waveguide structure multiple quantum well lasers,” IEEE J. Sel. Top. Quantum Electron. 21(6), 189–194 (2015). [CrossRef]
32. J.-P. Jansson, A. Mäntyniemi, and J. Kostamovaara, “A CMOS time-to-digital converter with better than 10 ps single-shot precision,” IEEE J. Solid-State Circuits 41(6), 1286–1296 (2006). [CrossRef]
33. J.-P. Jansson, A. Mäntyniemi, and J. Kostamovaara, “Synchronization in a multi-level CMOS time-to-digital Converter,” IEEE Trans. Circ. Syst. 56(8), 1622–1634 (2009). [CrossRef]
34. S. Jahromi, J. Jansson, I. Nissinen, J. Nissinen, and J. Kostamovaara, “A single chip laser radar receiver with a 9×9 SPAD detector array and a 10-channel TDC,” in European Solid-State Circuits Conference (2015), pp. 364–367.
35. J. Nissinen and J. Kostamovaara, “A 4 a peak current and 2 ns pulse width CMOS laser diode driver for high measurement rate applications,” in Proceedings of European Solid-State Circuits Conference (IEEE, 2013), pp. 355–358. [CrossRef]
36. J. H. Shapiro, “Photon-efficient 3D imaging,” in Applications of Lasers for Sensing and Free Space Communications (Optical Society of America, 2015), pp. LM3D–2.
37. D. Shin, A. Kirmani, V. K. Goyal, and J. H. Shapiro, “Photon-efficient computational 3-d and reflectivity imaging with single-photon detectors,” IEEE Trans. Computational Imaging 1(2), 112–125 (2015). [CrossRef]