High Dynamic Pixel Implementation Method for Adaptive Adjustment of Gate Voltage of Transfer Transistor on Chip

In CMOS Image Sensors (CIS), high dynamics are required to adapt to the variation of light intensity in different application scenarios. In this paper, we propose a highly dynamic method based on an 8T active pixel, which can adjust the gate voltage of the transfer transistor on the chip adaptively by changing the gate voltage of the transfer transistor in the pixel, to adjust the height of the potential barrier between the PD and FD nodes so that the PD node inside the pixel can fuse the signals of long and short exposures. The output of the sampled pixels is compared with the reference voltage, and the result is output to the gate of the transfer transistor through the DAC to determine the total exposure time allocation, achieving the effect of adaptive adjustment of the exposure time allocation by the pixel output. Based on the 55 nm process, the proposed method is verified by a specific circuit design, and the validation results show that the full trap capacity is 24 ke− and the dynamic range can be extended from 66.02 dB to 89.54 dB without decreasing the pixel fill factor and increasing the signal processing complexity, and the long and short exposure times can be adjusted adaptively according to the application scenario to obtain the optimal high dynamic. The solution is optimized for high dynamics.


Introduction
CMOS image sensors (CIS) have been increasingly developed in recent years because of low power consumption, large surface array, high performance, and integrability [1], which require image sensors with a sufficiently large dynamic range due to the existence of a very large range of variations in light intensity.The current methods to improve the dynamic range include the logarithmic output pixel technique [2], conditional reset technique, multiple exposure sampling technique [3], etc.These techniques expand the dynamic range to some extent, but the fixed pattern of logarithmic output pixel technique is very noisy and has the problem of poor pixel output consistency; conditional reset technique needs to add extra capacitance in the pixel and has low signal-to-noise ratio, which increases the complexity of the pixel; multiple exposure sampling technique needs to combine the image information captured by the image sensor twice or more in the on-chip memory according to a certain algorithm, which greatly increases the amount of hardware.
The dynamic range is defined as the ratio of the maximum unsaturated signal to the minimum measurable input signal.In this paper, based on 8T active pixels [4][5], we use the intra-pixel long and short exposure fusion technique [6][7], which achieves a segmented linear photoresponse curve by adjusting the transfer transistor gate pressure in the pixel to fuse the long and short exposure signals in one exposure directly, thus improving the method to achieve on-chip adaptive adjustment of transfer transistor gate pressure for highly dynamic pixel design.

Method for adjusting M TX gate voltage to achieve high dynamics
The 8T pixel structure and its operating timings are shown in Figure 1.First M RST and M TX reset PD and FD nodes and start the exposure.At the end of the exposure, the FD node is reset to clear the excess charge from the PD node during the exposure.At this time, the tail current tube M PC of M SF1 starts to work, M S1 and M S2 are on, and the reset signal is stored in C 2 , then M TX is turned on, M PC and M S1 continue to be on, and the signal level is stored in C 1 .Finally, M SEL turns on and reads out the reset signal stored in C 2 first, then M S2 turns on and reads out the signal level stored in C 1 .In the exposure method where the transfer transistor gate voltage is always kept low, the final full trap capacity at the PD node is: where Q PD_MAX is the full trap capacity of the PD point, V PDRST is the reset voltage value of the PD point, V PDMIN is the lowest voltage value that the PD point can reach in this state, t int is the total exposure time, P 0 is the maximum light intensity that can be sensed, and α is the photoelectric conversion factor.
We set in turn the M TX gate voltage V 1 , V 2 , and V 3 , as well as exposure time t 1 , t 2 , and t 3 , requiring V 1 >V 2 >V 3 and t 1 >t 2 >t 3 .In these states in turn after exposure, the lowest voltage that can be reached at the PD point are V PD_MIN1 , V PD_MIN2 , and V PD_MIN , and the maximum amount of charge that can be accommodated in the PD in each exposure period are: In this method, by setting the voltage and time of the transfer transistor during exposure, the compressed photoresponse curve as shown in Figure 2 1): where the main factors affecting P 1 are V PDMIN1 and t int /t 1 .Since t int /t 1 is approximately 1, the main factor affecting P 1 is the M TX gate voltage value for the first exposure.
The equation for the dynamic range expansion factor K is shown below: where V PD_MIN2 and t int /t 3 can adjust the dynamic range.The fusion of three exposure signals under different M TX gate voltages at the PD point within one exposure cycle of the pixel is a way to finally present three linear segments by using a compressed photoelectric conversion curve while enabling three different light ranges with different sensitivities.

Adaptive adjustment of M TX gate voltage on the chip
Adaptively adjusting the length of exposure time within an exposure cycle of the pixel requires the pixel to be able to read out while exposing and transmitting the signal level generated during exposure in real-time, and the pixel cannot be read out while exposing in the normal operating mode of the pixel.The circuit diagram is shown in Figure 3 (a), adding a column of sampled pixels outside the pixel array.Each sampled pixel cell exposure time isometric increases until it increases to the exposure time set in the pixel array, and the rest of the control signal is shifted back through the shift register to match the control signal of the M TX to achieve readout while exposure.The signal level needs to be sampled separately once after the M SEL conduction, and the output of the sampled pixels is shown in Figure 4.The final result of the DAC will be transmitted to the gate of the M TX in the pixel array, and the exposure process is divided into three segments of different lengths.
The structure of the sampling pixel is the same as that in the pixel array, but the photocurrent in the sampling pixel can be adjusted.As shown in Figure 4 (b), when the photocurrent value is small, the output slope of the sampling pixel is also small, which corresponds to a long exposure time and can better capture the dark details.When the photocurrent value is large, the output slope of the sampling pixel is large too, which corresponds to a corresponding increase in the short exposure time, and the charge in the PD node eventually reaches the full trapping capacity.In this way, the photocurrent value of the sampled pixel can be adjusted according to the usage scenario so that the gate voltage of the M TX can be adaptively adjusted to control the long and short exposure times.It was obtained by simulation that the best extended dynamic range is obtained by setting V 1 , V 2 , and V 3 to 1.4 V, 0.9 V, and 0.4 V, respectively.When the output voltage of the sampled pixel is greater than V ref1 , the output of Comparator 1 is 1, the input of DAC is 101, and the output is 1.4 V; when the output voltage of the sampled pixel is less than V ref1 , the output of Comparator 1 is 0, the input of DAC is 010, and the output of DAC is 0.9 V. V ref2 is set as 98% of the pixel swing when the output voltage of the sampled pixel is less than V ref2 , Comparator 2 output is 0, the input of the DAC is 001, and the output is 0.4 V. V ref3 is set as the lowest potential that the pixel output can reach.When the output voltage of the sampled pixel is less than V ref3 , Comparator 3 output is 0, at this time the input of the DAC is 000, and the output voltage is 0 V.
The output of the second comparator is also the control terminal of S 4 , and the output of the third comparator is also the control terminal of S 3 .The logic relationship between different outputs of different comparators is used to control the working timing of S 3 and S 4 , so that different DAC output voltage values are transferred to the gate voltage of the M TX at different times, and the working timing of the transferred reset level is controlled by controlling S 5 , the timing and final results are shown in Figure 5.

Validation results and analysis
In this paper, we implement the above method based on a 55 nm process with an 8T pixel structure and illustrate the results with an exposure time of 10 ms.As shown in Figure 6, V 1 = 2.5 V, V 1 = 2 V, and V 1 = 1.4 V. Three curves can prove the correctness of the theory that the value of the M TX voltage dominates in the effect on P 1 in Section 2, according to the curve obtained by setting V 1 = 1.4 V, V 2 = 0.9 V, and V 3 = 0.4 V. Three linear effect is most obvious.After determining the gate voltage, we change the distribution of exposure time, and fixed t 3 = 200 us is unchanged, under the premise of t 1 >t 2 >t 3 , and then we change t 1 from 4 ms to 9.5 ms.Fixed t 2 = 600 μs keeps unchanged, and we change t 3 from 100 μs to 300 μs.From Figure 6, it can be obtained that V 1 and t 1 mainly affect the position of P 1 , t 3 affects the final dynamic range expansion effect, and the slope of the third section t 3 /t int is inversely proportional, but at the same time, the slope of the third section of the fold is too small, which will lead to the small third section of the sensitivity and the accumulation of a charge by the change in voltage value, so the selection of t 3 needs to be considered in compromise.Figure 6.t 1 and V 1 are changed when t 3 =200 μs Figure 7. t 1 is changed when t 2 =600 μs The curves in Figure 6 and Figure 7 can be obtained when t 1 occupies 90%-92% of the total exposure time, and t 3 accounts for 1%-1.5% of the total exposure time when the three linear effect is extended and dynamic range effect is best.Under the premise that the photoelectric conversion is linear, the ratio of V ref to pixel full swing is set according to the same ratio.As shown in Figure 8, we set the exposure time of the sampled pixel by the 100 μs pitch increase.The output of the sampled pixel is shown, through the dynamic latching comparator, to compare s_pixel with the reference voltage value to obtain the output results of the DAC, DAC 2 , and DAC 3 .The comparator results control the timing of Switches S 3 , S 4 , and S 5 .In the case of the maximum photocurrent that can be sensed in conventional exposure mode, the simulation yields pixel cell readout results.It can be seen that at the original maximum photocurrent, the output of the pixel at this time has not yet reached the minimum value.Scanning photocurrent is to obtain the extended dynamic range results as shown in Figure 9, calculated by Equation ( 7) to obtain the dynamic range from the original 66.02 dB extended to 89.54 dB.
Table 1 shows a comparative analysis between the circuit designed in this paper and the high dynamic pixel method used in [8][9][10].It shows that the circuit in this paper has the advantage of extending the dynamic range while adaptively adjusting the M TX gate voltage according to the illumination to achieve a large dynamic range and different light sensitivity for different illumination ranges.a Note: "-" indicates no discovery.

Conclusions
This paper achieves a dynamic range of 89.54 dB by an on-chip adaptive adjustment of the M TX gate voltage, enabling the fusion of short and long exposure signals within one exposure cycle of a pixel without increasing the complexity of the pixel structure and without the need for complex digital signal processing off-chip and readout in one frame time, which saves time, area, and power consumption.

Figure 1 .
Figure 1.The 8T pixel structure and working timings

Figure 2 .
Figure 2. Schematic diagram of the compressed photoelectric response curve

Figure 3
Figure 3 (a).Pixel array Figure 3 (b).Circuit for output control DAC As shown in Figure 3 (b), the circuit diagram will be sampled pixel output through three comparators and reference voltage for real-time comparison.The comparators used in the circuit are dynamic latching comparators, and different comparison results through the DAC will produce different voltage values.The final result of the DAC will be transmitted to the gate of the M TX in the pixel array, and the exposure process is divided into three segments of different lengths.The structure of the sampling pixel is the same as that in the pixel array, but the photocurrent in the sampling pixel can be adjusted.As shown in Figure4(b), when the photocurrent value is small, the output slope of the sampling pixel is also small, which corresponds to a long exposure time and can better capture the dark details.When the photocurrent value is large, the output slope of the sampling

Figure 4
Figure 4 (a).Working timing of sampling pixel Figure 4 (b).Adjustment of M TX gate voltage

Figure 8 .
Figure 8. Final timing and results Figure 9. Final results

Table 1 .
Literature comparison