A 134×132 4-Tap CMOS Indirect Time-of-Flight Range Imager Using In-Pixel Memory Array With 10 Kfps High-Speed Mode and High Precision Mode

This article presents a prototype 4-tap indirect time-of-flight (iToF) CMOS imager with high-speed (HS) range imaging capacity. The sensor features an HS charge modulator, in-pixel memory array, and sub-frame ToF operation, enabling up to 10 Kfps range imaging with eight recording frames and < 1.82% depth noise for 0.3–1.4 m range under HS mode. This sensor also operates in high-precision (HP) mode, achieving < 1.77% depth noise for the 0.4–5.4 m range at 90 frames/s by averaging the subframe signals. With a pixel size of <inline-formula> <tex-math notation="LaTeX">$22.4^{H} \times 16^{V}\,\,\mu \text{m}$ </tex-math></inline-formula> and <inline-formula> <tex-math notation="LaTeX">$134^{H} \times 132^{V}\,\,4$ </tex-math></inline-formula>-tap pixel array, the sensor successfully demonstrated precise depth imaging under HP mode and clear 3-D imaging for rapid motion objects under HS mode. The potential application of the sensor and future improvements are also discussed.

Abstract-This article presents a prototype 4-tap indirect time-of-flight (iToF) CMOS imager with high-speed (HS) range imaging capacity.The sensor features an HS charge modulator, in-pixel memory array, and sub-frame ToF operation, enabling up to 10 Kfps range imaging with eight recording frames and <1.82% depth noise for 0.3-1.4m range under HS mode.This sensor also operates in high-precision (HP) mode, achieving <1.77% depth noise for the 0.4-5.4m range at 90 frames/s by averaging the subframe signals.With a pixel size of 22.4 H × 16 V µm and 134 H × 132 V 4-tap pixel array, the sensor successfully demonstrated precise depth imaging under HP mode and clear 3-D imaging for rapid motion objects under HS mode.The potential application of the sensor and future improvements are also discussed.

I. INTRODUCTION
T HE growing demand for 3-D sensing in applications such as modeling, biometrics, gesture control, and virtual reality (VR)/augmented reality (AR) has led to the development of 3-D imaging technologies with higher spatial resolution and precision.In recent years, new possibilities have emerged in the field of computer vision, including robotics, industrial automation, autonomous vehicles, and medical assistance systems.To enhance the reliability of scene analysis and decision-making for tasks such as object recognition, tracking, and navigation, depth images with higher temporal resolution are required.Moreover, achieving a performance level over 1 Kfps would enable more opportunities in advanced machine vision applications [1].Therefore, there is a desire to develop high-speed (HS) range imagers that greatly exceed standard video rates (30/60 frames/s) to provide crucial spatial information.
A structured light algorithm as well as a direct time-offlight (dToF) SPAD sensor has been reported for HS 3-D imaging, achieving up to 1 Kfps [1], [2].However, the need for GPU acceleration in complex post-processing and HS readout circuits limits the feasibility of higher frame rates and image resolutions.Besides, larger module sizes and higher power consumption reduce the portability of the sensor system.
In an iToF system, it has been demonstrated that the signalto-noise ratio (SNR), modulation frequency, and demodulation contrast (DC) are inversely proportional to the depth noise [9].Therefore, designs with higher full-well capacity (FWC) are often preferred to achieve higher system SNR and ambient light tolerance [10], [11].In addition, a smaller pixel pitch is also desired to improve DC when increasing the modulation frequency.However, these limit the increase in frame speed due to the longer exposure period and multi-frequency synthesis demands [12].Consequently, current iToF imaging system designs need to compromise between these trade-offs, resulting in a confined field of applications.
Recently, a 4-tap iToF range imager enables low noise imaging at 90 frames/s and HS imaging at 2 K to 10 Kfps using an in-pixel memory array and sub-frame ToF operation was reported [13].This sensor exhibits the potential for achieving fine depth image quality while also having the capacity to provide high temporal resolution 3-D images.
In this article, we discuss in detail the design concepts, theoretical calculations, and measured characteristics of the developed CIS for range imaging.In addition, we demonstrate the potential application of combining the high precision and HS depth imaging.
Section II describes key technologies, design concepts, and system structure of this work.Section III shows measurement results, demonstration images, and discussion.Finally, the conclusion is given in Section IV.

A. Indirect-ToF Using 4-Tap Short Pulse Modulation
To address the challenges of HS imaging, multi-tap architectures have emerged as a suitable scheme to mitigate the motion artifact and enhance the depth precision [6], [14].The 4-tap 4-phase iToF operation using sinusoidally modulated light is commonly adopted for continuous wave (CW) modulation due to its robustness and stability toward a higher modulation frequency.Furthermore, multi-frequency synthesis can be applied to extend the unambiguous range [5], [7].
In contrast, the short pulse (SP) modulation prepared by 4-tap 1-drain iToF pixel has been reported to have better ambient light tolerance due to the concentrated light energy from the emitter [15], [16].In addition, the range-shifted technique can be adopted to increase the number of timewindows (TWs) for long-range detection [11], [17].
Fig. 1 illustrates the timing diagram of a 4-tap iToF operation using CW and SP modulation under the same emitter power condition and modulation frequency ( f m ).The modulation cycle (T C ) is equally divided by four sampling clocks (T P ).The reflected signals, including the modulated light and background light, are demodulated into floating diffusions (FDs) by Tap1, Tap2, Tap3, and Tap4, and are denoted by Q 1 , Q 2 , Q 3 , and Q 4 , respectively.The distance d CW and d SP , with background light cancellation, can be calculated using the following equations, assuming the ideal sinusoidal and square pulse lights are given: where c is the speed of light.Note that (2) is formulated for the distance range where the phase shift (ϕ) is within TW1.Then, by applying the error propagation to the distance equations, the depth noise, σ CW and σ SP , can be expressed by the following equations [5] and [14]: where DC is the DC and R S = N ϕ /N S .The number of total electrons integrated in a unit pixel from the modulated light is denoted by N S .The number of electrons generated by background light in a unit tap is denoted by N B , and the total electrons is denoted by N BT .The FD input referred readout noise (RN), which contains the pixel transistors and readout circuit, is denoted by RN.Under SP modulation, the number of signal electrons in Tap2 is proportional to the phase shift (ϕ), and it is denoted by N ϕ .
To compare the performance of iToF operation with CW and SP modulation, we analyzed the system SNR and theoretical depth noise at different RN levels.The detection range was set to 0.4-6 m with T p of 10 ns and T c of 40 ns.We assumed no ambient light (N BT = 0) and assigned N S of 20 000 at 0.4 m.The DC of both modulation methods was set to 90%, which is an achievable value in the state-of-art iToF system.The SNR is calculated by the following equation: Fig. 2(a) and (b) shows that SP modulation can provide better depth noise, particularity when the SNR is dominated by signal shot noise.However, it is important to note that in a scenario with higher RN, the SNR degrades more rapidly, leading to a diminished advantage when using SP modulation.
Therefore, to achieve lower depth noise and higher frame rates, a desirable solution is the development of a 4-tap iToF image sensor using SP modulation with enhanced SNR.In this work, to minimize the image processing effort and maximize the frame speed, a single frequency SP iToF system is employed without drain gate and range-shift technique.
Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.

B. Sub-Frame ToF Operation
To improve the system SNR of iToF imagers, increasing the FWC and exposure time are common strategies, but they come at the expense of higher input referred noise due to a lower conversion gain (CG) caused by higher FD capacitance.In contrast, pixel binning techniques, such as charge domain, analog domain, and digital domain binning can also be useful in reducing the RN but at the cost of reducing spatial resolution [18].However, these approaches require longer frame-toframe latency while increasing the pixel counts, making them unsuitable for HS imaging.
In this work, we introduce the sub-frame ToF operation with an in-pixel memory array to enhance the SNR while keeping the capacity of high temporal resolution imaging.As shown in Fig. 3(a), the full modulation period is divided into subframes, and the signals are sampled into individual memory cells.Subsequently, by applying different readout timings, as shown in Fig. 3(b), two types of imaging modes can be performed.
1) HP Mode: In HP mode, the modulated subframes are combined using charge-domain binning by mixing the signal charges in the memories.Owing to the averaging effect, the noise can be reduced including shot noise, flicker/thermal noise of pixel transistors, and kTC noise of the in-pixel memory.The theoretical SNR is expressed by the following equation: where N SubF is the number of subframes, CG is the CG, and N PIX , N MEM , and N CKTs represent the number of noise electrons from pixel, in-pixel memory, and other readout circuits, respectively.These can be obtained by dividing the FD referred noise voltage σ PIX , σ MEM , and σ CKTs by CG.
The calculated SNR characteristic is shown in Fig. 4(a), where a reference is established using the conventional operation with a CG of 10 µV/e − for comparison with the subframe operation.A constant CG/N SubF (µV/e − ) of 10 and N S •N SubF (e − ) of 100 000 are used to have an equal amount of signal electrons under a certain exposure condition.Here, σ PIX of 300 µV, σ MEM of 370 µV, and σ CKTs of 200 µV are assumed.
Fig. 4(b) depicts the SNR enhancement achieved through HP mode.Minor improvement was observed in shorter distances due to shot noise dominance.As the distance increases, a higher number of subframes with higher CG results in better SNR due to the reduction of equivalent noise electrons.The optimal number of subframes is eight, considering SNR enhancement and hardware complexity.
2) HS Mode: HS mode readout the memory cells individually, enabling the acquisition of high temporal resolution images.In this burst imaging mode, the frame rate is defined as one divided by the subframe period, and the available record length is determined by the number of subframes.The SNR is expressed by the following equation:

C. Circuit Architecture and Operation
In this prototype sensor, a 4-tap iToF pixel was developed with eight subframes.To reduce the sampling period and minimize the exposure deadtime between subframes, an inpixel memory array was employed.Fig. 5(a) and (b) shows the diagrams of the pixel, which consists of a 4-tap modulator with HS charge collection photodiode (PD), demodulation gates (TG1-TG4), four sets of buried channel source follower (PSF), current source (PCS) with cascode switch (CSC), auto-zeroing capacitor (C AZ ), and 4 × 8 1-T 1-C analog memory array with control devices [memory write (MW), memory reset (MRST),  [19], [20], providing capacitance of 30 and 170 fF, respectively.
To reduce the subframe period and ensure good DC, the PD was designed with the following concepts: 1) increasing the size of PD to increase the number of generated charges; 2) enlarging the n-layer region to enhance the charge collection efficiency; 3) creating a dopant concentration ladder to generate an electric field toward the center of modulator [21]; and 4) implementing a compact 4-tap structure to improve the electron sorting efficiency by utilizing fringe electric field [22].
In this work, we also utilize a spiky triangular-shaped layout [23], [24] with three levels of n-layer (N1-N3) to create a linear potential gradient along the charge transfer path.This ensures that the generated electrons can be driven efficiently regardless of their initial position.
Simulated potential diagrams of electron transport to FD1 and FD2 are shown in Fig. 6(a) and (b), respectively.In the simulation, electrons were initially placed at a depth of 1 µm from the far end of the PD.Fig. 6(c) provides a comparison of their transfer paths.To reduce power consumption, the TG-ON voltage of 2.4 V and TG-OFF voltage of 0.4 V were used while maintaining good electron sorting efficiency.The HS charge modulator can collect charges within 0.8 ns owing to an electric field of over 1200 V/cm across the PD to the farthest FD, FD2, and FD4.C MEM , and C AZ are reset.Auto-zeroing is then performed by PR and PH in sequence (t2, t3) to eliminate the thermal noise and offset from FD reset.After the modulation is completed, the 4-tap signals are simultaneously sampled into the corresponding memories by MS (t4).The auto-zeroing operation reduces the required number of voltage samplings by half, resulting in a smaller memory array.It is worth noting that during the modulation, PCS is turned off by CSC to reduce the power consumption.
To optimize the HP and HS modes, different modulation types and readout timing are applied, as shown in Fig. 7(a) and (b), respectively.
1) HP Mode: In HP mode, a conventional 4-tap iToF modulation (LED HP ) is applied to extend the unambiguous range.In addition, to achieve higher SNR for lower depth noise, the signal charges in the 8-memory of each tap are summed up before the column sample/hold (S/H) operation.
2) HS Mode: In HS mode, a pseudo-2-tap iToF modulation (LED HS ) is employed to increase the system SNR, despite having a shorter detection range.This is accomplished by doubling the light pulse frequency, which is demodulated alternatively by TG1/TG2 and TG3/TG4.Consequently, the selected memories in PH1 and PH3 and PH2 and PH4 are mixed before the column S/H, completing the pseudo-2-tap operation.Finally, the memory signal of each subframe is readout in sequence to obtain burst images.
For both modes, the delta-double sampling (DDS) readout [25] is performed at voltage of memory output (VMO) to reduce both fixed pattern noise and low frequency noise due to MSF readout.× 16 V µm.The demodulation clocks (TGs), generated by shift registers, are driven by two sets of both top and bottom sides of the pixel array to minimize TG pulse distortion under high-frequency modulation.The LED trigger and delay control circuits are implemented to fine-tune the edge of modulation light pulse.During the rolling readout, the signals in the pixel memory arrays (MEM) are sampled by the column SH (CSH) and selected by horizontal shift register (HSR).Finally, the analog output signals from the output buffer (OB) are quantized by an off-chip 14-bits ADC.

III. MEASUREMENT RESULTS
Fig. 9 shows the micrograph of the developed prototype chip.It was fabricated using a 0.18-µm 1-poly-Si 5-metal CIS process technology with 8-µm-thick P-epi on N-sub wafer.The power supply voltage is 3.3/2.8/2.4V for analog/digital/TG circuits, respectively.The die size is 4.8 mm H × 4.8 mm V .In this prototype sensor, a CG of 84.8µV/e − , and an FWC around 12ke − were confirmed at FD1.To evaluate the proposed HP and HS mode, conventional operation with high CG and low CG, denoted by HCG and LCG, respectively, are used for comparison.The basic characteristics are listed in Table I.Note that the results of LCG were calculated by a factor of 8.The LCG exhibits an advantage over HCG in higher illuminance conditions, where shot noise dominates the SNR, due to its higher FWC.In contrast, the HP mode provided over 8.4 dB SNR enhancement across the measured illuminance range, contributing to better depth noise performance in iToF range imaging.In HP mode, the total FD input-referred noise was 407 µV, consisting of the noise from pixel circuits (254 µV) and readout circuits (318 µV).
The DC characteristic of the proposed 4-tap iToF sensor with SP modulation has been evaluated and plotted in Fig. 11 with different demodulation pulse widths (T P ), and was calculated by the following equation [16]: where the demodulated signal in each tap is denoted by S n .A 4-tap averaged DC of 78.7% at 10 ns demodulation pulse was obtained by the averaged data from 10 × 10 pixels in the center of the array.However, there is a DC difference of around 14% between Tap1/3 and Tap2/4 due to the asymmetrical structure of the charge modulator.The DC is strongly related to the charge collection efficiency of each Tap.Tap2/4 requires a longer electron transfer time due to a weaker electric field with longer transfer path, resulting in a lower DC.Hence, assuming that the transfer time difference remains constant, decreasing the T P exacerbates the DC difference between Taps.
The depth performance was evaluated by analyzing the central 100 pixels over 100 consecutive frames.The system was set up with a 90% reflectivity white flat board as the target, and an 850 nm vertical-cavity surface-emitting laser (VCSEL) with a peak power of 8-W generating the modulation light pulse with a T P of 10 ns.An F/1.4 lens with an IR bandpass filter was used, and the measurement was conducted indoors with <500 lx fluorescent light.The detailed system parameters are summarized in Table II.
The modulated light and TG pulses were set to 10 ns, which allowed for an unambiguous range of 1.5 m with a single TW.The HP operation provides 4-TW with the cycle time (T C ) of 40 ns, whereas the HS mode has only 1-TW due to the pseudo-2-tap operation.A time delay of ∼1 ns existed between TG1 pulse and the light pulse.The slope and time offset of the measured depth were calibrated for each TW [26].

A. HP Mode
The theoretical depth noise of the proposed HP mode was compared.The equation can be expressed by the following equation: where the measured RN is denoted by RN HP .Here, f m of 25 MHz, N B of 0, RN HP of 4.8, N SubF of 8, DC of 0.86 for TW2/4, and 0.72 for TW1/3 were applied.At the distance of 0.4 m, N S of 20 000 was used, with 12 000 electrons in Tap1 and 8000 in Tap2.Fig. 12(a) and (b) shows the measured results of depth accuracy and depth noise, respectively.The system is capable of measuring distances ranging from 0.4 to 5.8 m at a 90 frames/s framerate with an exposure time of 1.0 ms per subframe.The measured depth nonlinearity was <1.62%, which was attributed to tap mismatches, non-ideal distortion caused by the limited bandwidth of the light pulse, and reflected stray light from the measurement system.The depth noise was measured to be <1.77% for the range within 5.4 m.Compared to the conventional HCG and LCG operation, the proposed HP mode provides a better depth noise performance across all ranges.The captured sample images and environment setup are depicted in Fig. 13.The nearest object was placed at 45 cm, and an alphabet "U " was positioned at 2.55 m, which was 15 cm in front of the background panel.While the intensity map demonstrated a rapid decrease in reflected light as the distance increased, the proposed HP mode still delivered fine depth resolution and a distinct image of "U " without requiring frame averaging.The depth error observed at the edge region surrounding the objects was attributed to the "flying pixel" effect caused by spatial sampling issues [27].

B. HS Mode
The equation of HS mode, which was calculated with pseudo-2-tap operation, can be expressed by the following equation: where the measured RN is denoted by RN HS .Here, f m of 50 MHz, N S of 16400, N B of 0, RN HS of 7.9, and DC of 0.72 were applied.Fig. 15 demonstrates depth imaging at 2 Kfps, where a spinning alphabet "T " at ∼4500 rpm blocked the objects behind it.By using the HS mode, the captured frame could be disassembled into eight burst frames with a temporal resolution of 0.5 cms, enabling clear depth images of the background objects and observing target.Note that the burst images in HS mode are readout and refreshed every 33.3 ms.Sample images with a higher speed of 5 and 10 Kfps are to be shown in Fig. 16.
The calculated noise curves for both HP and HS modes provide a reliable estimate of the expected depth noise based on the measured system parameters.However, it should be noted that in practical iToF measurement, the depth performance is affected by several system limitations and uncertainties, such as clock distortion and jitter, limited bandwidth of the modulated light pulse for SP modulation, unstable sensor and emitter power, and heat buildup during the sensor operation.To obtain relatively stable measurement results, external voltage sources and ADC were used for this prototype sensor.Meanwhile, a heat sink was attached behind the designed emitter module.
In real practice, this development is suitable for constructing 3-D environments for human behavior analysis using HP mode, thanks to its low depth noise, suppressed Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.motion artifact, and wide-range imaging capability.In contrast, HS mode provides high temporal resolution images that can be applied in machine vision for rapid motion recognition and analysis.Furthermore, the combination of HP and HS mode has a great potential to excel in event-driven applications, for example, a comprehensive automotive safety system.
In regular circumstances, the HP mode can continuously monitor the driver's behavior for gesture control and alert them in case of any signs of impairment, while the HS mode can be activated in case of any sudden impact or collision.In the event of severe accidents, where the deceleration forces can exceed 50G and cause the human body to come to a sudden halt within a couple milliseconds.At this moment, an HS burst imaging is needed to record the crucial spatial information.
In the experiment shown in Fig. 16, the accident was indicated by a bouncing plate using rubber bands, which was captured under 10 and 5 Kfps, with record durations of 0.8 and 1.5 ms, respectively.The pattern on the target, and its displacement and rotation, were all successfully observed.
The captured burst images can then be used to analyze the rapid motion and detect any potential injuries that might have occurred.In addition, the event-driven HS mode can quickly capture and analyze the relevant data without requiring continuous operation at such high framerate, reducing power consumption, and extending the system lifespan.
Table III compares the performance of the developed prototype sensor to state-of-the-art iToF sensors.The figure-of-merit (FoM) can be calculated using the following equation, which is another expression of [7, (3)]: where the depth noise (%) is the maximum value measured in the selected range, and pixel rate (PR) is calculated using (12).In the HS mode, burst images are refreshed with a period of 33.3 ms.Hence, the equivalent PR is expressed in the following equation: PR HS = N SubF × pixel count refresh period pixel s .
In an iToF ranging system, the modulation period is adjusted based on the signal saturation level at the minimum distance.However, the power of the modulated light decreases rapidly with depth, limiting the maximum distance due to low SNR.Hence, the detection range of the measured data should be taken into consideration.For comparison, a range-FoM (R-FoM) is defined as in the following equation: where range ratio is (Distance MAX /Distance MIN ) of the selected range, and the Distance MAX is chosen based on the minimum achievable R-FoM, which indicates the optimal working range with the highest efficiency for an iToF imager.This prototype sensor achieves an R-FoM of 16 and 18.4 pJ/pixel in HP and HS mode, respectively, for a range of 0.4-5.4 and 0.3-1.4 m.
For future development, the performances can be improved from several aspects.First, sensor resolution can be significantly enhanced by introducing the backside illumination  (BSI) 3-D stacking technologies.These would improve layout flexibility and area efficiency to achieve higher fillfactor.Second, IR-responsivity enhancing technologies such as microlens, deep trench isolation (DTI), and pyramid surfaces for diffraction (PSD) structure [28] can be adopted to reduce the exposure time, resulting in a higher frame speed and lower power consumption.Third, to achieve better DC, creating doping gradient on P-epi [24] is beneficial to increasing charge collection efficiency in the vertical direction, resulting in better depth precision.Lastly, the RN can be reduced by implementing a column-ADC circuit and using higher density in-pixel memory with textured deep trench SiN capacitors [29] or high-capacity DRAM capacitors [30], [31].Moreover, by utilizing high-density capacitor and 3-D stacking techniques, it is possible to implement more in-pixel memory, which can extend the duration of burst imaging even at a higher frame rate.Note that the frame time of HS mode is only limited by MRST and sampling time, which is less than 1 µs in this work.
These advancements in sensor technology can enable higher resolution, lower power consumption, faster frame rate, and better ranging performance for the development of HS and high precision iToF depth imager.

IV. CONCLUSION
An iToF range image sensor with in-pixel analog memory array and sub-frame ToF operation was developed and demonstrated the unprecedented HS range imaging as well as the depth precision enhancement.The fabricated 22.4 H × 16 V µm pitch, 134 H × 132 V 4-tap pixel iToF imager exhibited up to 10 Kfps range imaging with the HS mode and <1.77% depth noise for 0.4-5.4m range with the HP mode.This development opens a new avenue for HS 3-D imaging applications for machine vision and more.

Fig. 2 .
Fig. 2. (a) System noise with SNR and (b) theoretical depth noise over distance with different RN values.

Fig. 3 .
Fig. 3. Operation principle of (a) sub-frame ToF modulation and (b) HP and HS mode readout.

Fig. 7 (
Fig.7(a) shows the detailed timing diagram of the subframe operation, constructed using eight subframes with SP modulation.At the beginning of each subframe (t1), the FDs,

Fig. 10
Fig.9shows the micrograph of the developed prototype chip.It was fabricated using a 0.18-µm 1-poly-Si 5-metal CIS process technology with 8-µm-thick P-epi on N-sub wafer.The power supply voltage is 3.3/2.8/2.4V for analog/digital/TG circuits, respectively.The die size is 4.8 mm H × 4.8 mm V .In this prototype sensor, a CG of 84.8µV/e − , and an FWC around 12ke − were confirmed at FD1.To evaluate the proposed HP and HS mode, conventional operation with high CG and low CG, denoted by HCG and LCG, respectively, are used for comparison.The basic characteristics are listed in TableI.Note that the results of LCG were calculated by a factor of 8. Fig.10compares the SNR characteristic of HCG, LCG, and HP over different illuminance levels, indicating the maximum achievable SNR at different distances (d) in an iToF system.

Fig. 11 .
Fig. 11.DC measurement results of each tap with different demodulation pulse widths (T P ).

Fig. 14 (
Fig. 14(a) and (b) shows the depth measurement results obtained at 2 Kfps with 0.5-ms exposure time.The shorter exposure time allows for a closer measurable range of 0.3 m without signal saturation.The depth performance of each subframe was characterized separately for the range of 0.3 to 1.4 m.The 8-subframe averaged depth nonlinearity was <1.91%, with depth noise <1.82%.Fig.15demonstratesdepth imaging at 2 Kfps, where a spinning alphabet "T " at ∼4500 rpm blocked the objects behind it.By using the HS mode, the captured frame could be disassembled into eight burst frames with a temporal resolution of 0.5 cms, enabling clear depth images of the background objects and observing target.Note that the burst images in HS mode are readout and refreshed every 33.3 ms.Sample images with a higher speed of 5 and 10 Kfps are to be shown in Fig.16.The calculated noise curves for both HP and HS modes provide a reliable estimate of the expected depth noise based on the measured system parameters.However, it should be noted that in practical iToF measurement, the depth performance is affected by several system limitations and uncertainties, such as clock distortion and jitter, limited bandwidth of the modulated light pulse for SP modulation, unstable sensor and emitter power, and heat buildup during the sensor operation.To obtain relatively stable measurement results, external voltage sources and ADC were used for this prototype sensor.Meanwhile, a heat sink was attached behind the designed emitter module.In real practice, this development is suitable for constructing 3-D environments for human behavior analysis using HP mode, thanks to its low depth noise, suppressed

Fig. 16 .
Fig. 16.Demonstration of a scenario for the automotive safety system using the proposed sensor with (a) HP mode at 90 frames/s, (b) HS mode at 10 Kfps, and (c) HS mode at 5 Kfps.
A 134 × 132 4-Tap CMOS Indirect Time-of-Flight Range Imager Using In-Pixel Memory Array With 10 Kfps High-Speed Mode and High Precision Mode Chia-Chi Kuo , Student Member, IEEE, and Rihito Kuroda , Member, IEEE

TABLE I CHARACTERISTIC
SUMMARY OF MODES Fig. 8 the functional block diagram of the proposed iToF imager, which consists of a 134 H × 132 V pixel array with a pixel size of 22.4 H

TABLE III PERFORMANCES
SUMMARY AND COMPARISON