2.1. Traditional Structure
The structure of traditional Single Slope ADC is shown in
Figure 1, which comprises a comparator, ramp generator, and counter.
Vpixel is the output signal of the pixel in the CMOS image sensor. The ramp generator initiates a gradually decreasing voltage signal, with the counter tallying from the onset of this ramp. The pixel output is then compared to this slope voltage, yielding the SS ADC’s quantization result after decoding.
The TDC-based SS ADC converts the counter into a TDC. In this mode of operation, the SS ADC compares the pixel signal with the ramp signal. If the pixel signal voltage exceeds the ramp voltage, the comparator flips, and its output is transmitted to the TDC. The TDC then measures the time from the comparator flip to the end of the clock cycle.
The traditional TDC structure for CMOS image sensors is shown in
Figure 2, which is mainly composed of a delayed phase-locked loop, delay chain, and D flip-flop.
DFF_in<
N> represents the input of the
N-th D flip-flop, which is the output of the
N-th delay unit.
Comp_outb represents the inversion signal of the comparator. The pixel output is transmitted to the column level for comparison with the reference ramp, and the point in time at which the comparator flips is transmitted to the delay-locked loop (DLL) delay chain. In order to complete
n-bit quantization, 2
n delay units are required. The rising edge of the comparator flip is delayed by each delay unit with the same time accuracy, and the output of each delay unit is connected to the signal input of the D flip-flop. The rising edge of the next clock cycle will be sent to the clock input of the D flip-flop as a
stop signal. The clock signal will sample the signal input from the D flip-flop and read the output of the delay unit into the D flip-flop. Then the D flip-flop is used as a shift register, the data in the D flip-flop are passed to the decoder, and the final
N-bit quantization result is obtained.
2.2. Power Consumption Analysis of TDC
The main power consumption of a TDC comes from the power consumption of the delay unit and the D flip-flop, where the delay unit ensures the accuracy of the delay time, and its current magnitude is determined by the
Vc provided by the DLL. The power consumption of the CMOS circuit is
where
PT is the total power consumption,
Pst is the static power consumption, and
Pdyn is the dynamic power consumption. Static power consumption is mainly caused by a drain current, and static power consumption is
where,
ID is the drain current and
VDD is the power supply voltage. Dynamic power consumption is mainly caused by the charge and discharge of the circuit to the node capacitor. Each charge (or discharge) of a node capacitor
C will lead to the dynamic power consumption of 1/2
CVDD2. Therefore, the average static power consumption of a node
i in the circuit during the whole working time can be expressed as [
13]
where
fclk is the clock frequency and
Esw is the switching activity, which represents the average number of jumps in the node
i signal in each clock cycle. Therefore, the more the number of signal jumps in the D flip-flop, the higher its dynamic power consumption, resulting in a higher total power consumption.
In the simulation process, it can be found that when the signal input of the D flip-flop is kept at a low level, its power consumption level is far less than that when the signal input has a high level. Simulation of the D flip-flop was carried out. The simulation results are as shown in
Figure 3, respectively. The current condition of the output in the D flip-flop becomes high-level and the current condition of the input and output of the D flip-flop is kept low-level. The average current when the output remained low was 32.88 nA. The average current when the output changed was 217.2 nA. It could be seen that the average current in D flip-flop would be significantly increased by flipping. The power consumption of D flip-flop was mainly the power consumption of its flipping. Therefore, if an advanced counting method is adopted to reduce the flipping times of the D flip-flop in TDC, low power consumption can be achieved.
The output of the delay chain in the quantization process is shown in
Figure 4. Where
Comp_outb1 and
Comp_outb2 represent the inversion signal of the comparator in the first and second rows, respectively. When the
stop signal arrives, each D flip-flop will sample the output of the delay unit and generate a thermometer code with one and zero separated as 111…00. Due to the continuity of the analog signal, the light intensity information between the two adjacent rows of pixels is generally very close, so its quantization code value will also be very close. If a more simplified quantization method that quantizes the difference between two rows of pixel signals rather than the full signal can be adopted to eliminate the quantization of the same part between the two rows of signals, the quantification of redundant information will be removed. The number of output signals of one can be reduced; in other words, the number of flips of the D flip-flop can be reduced, which could reduce the counting power consumption.
The main source of ADC differential nonlinearity is the random variation in the delay time of the delay unit buffer caused by the inconsistency and mismatch of the device manufacturing. The greater the number of delay buffers, the greater the differential nonlinearity, which will cause a certain error. In addition, the D flip-flop must ensure that it has a narrow metastable width, which must be shorter than the delay period of the delay unit so as to reduce the TDC time measurement error.
2.3. Proposed TDC
The improved column-level readout architecture is shown in
Figure 5, where
start1 and
stop1 are the start and stop signals of the odd-line TDC, respectively,
start2 and
stop2 are the start and stop signals of the even-line TDC, respectively, and OR1 and OR2 are the outputs of the two OR gates, respectively. Since the output of two rows of pixels needs to be compared to quantify their difference, two column parallel buses need to be used for a simultaneous readout, and the output of pixels is sent to the column-level comparator for comparison with the reference slope voltage, respectively. When the output signals of the two rows of pixels are different, the flip time of the comparator will also be different. By sending the outputs of the two comparators into two sets of TDC for quantization at the same time, the flip time point of the comparator, i.e., the output signal of the pixel, can be encoded.
In high-speed image sensors, considering the RC delay in the row driver signal transmission line and column bus, the row driver signal is difficult to be established from the timing control driver circuit on both sides of the pixel array to the middlemost column pixel, and the pixel signal is difficult to be output from the middlemost row to the column level through the column bus during the row selection time.
Under the premise of not reducing the pixel array and frame rate, to achieve complete signal transmission within the row time, the mode of simultaneous readout of two column buses proposed in this work can be adopted. Each column pixel shares two column buses, and the pixel output of odd and even rows is transmitted to the column level, respectively. In this way, the timing control driver circuit can select two adjacent rows of pixels each time. The signal is selected and read at the same time; each column bus only needs to complete the selecting and reading of half of the pixel output in one frame time, and the row selection time can be doubled, which greatly alleviates the pressure of signal establishment in the row selection time. Therefore, the proposed TDC is very suitable for high-speed CMOS image sensors.
For TDC to quantify the difference between the output of two rows of pixels, it is necessary to quantify the time difference between the flip time points of the two rows of comparators. A series of logic circuits were designed before TDC, and the working sequence is shown in
Figure 6, where
Comp_out1b is the inverting signal for the odd-line comparator,
Comp_out2b is the inverting signal for the even-line comparator, and T1 and T2 represent the length of time that the TDC of odd and even lines need to be quantized, respectively.
In the initial state disjunction gate output, OR1 is high when Comp_out1, i.e., the first-row comparator, flip time point arrives earlier than Comp_out2, disjunction gate output remains high, the output of multiplexer A is the delay in the first-row comparator flip invert signal, and the signal which will be delayed by the delayed chain is sent to the start signal input of the first row TDC. At this time, the clock input of D flip-flop has not yet arrived, and its output is the default value after reset, i.e., the low level. The output of multiplexer B is also the delay in the first row in the comparator flipping invert signal, and the start signal of the two rows of TDC is the delay in the first row in the comparator flipping invert signal. After that, even if the second row in the comparator flipping time arrives, the value of OR1 remains unchanged. The output of the D flip-flop will also remain low; Since the output of the D flip-flop remains low, the stop signal of the first-row TDC is the rising edge of the next clock cycle, and the stop signal of the second-row TDC is the delay in the second-row comparator flip invert signal.
When Comp_out2 arrives before Comp_out1, the output OR1 of the disjunction gate will become low, the output of multiplexer A will be the delay in the second-row comparator flip invert signal, the output of the D flip-flop will also jump to a high level, and multiplexer B will select the delay in the second-row comparator flip invert signal. At this time, the stop signal of the second row in the TDC is the rising edge of the next clock cycle, and the stop signal of the first row in the TDC is the delay in the first row in the comparator flip invert signal.
It can be seen that the first arrived comparator flip signal will be fully quantized from the time of the flip time point to the rising edge of the next clock cycle, and the other row in the TDC completes the quantization of the time difference between the flip points of the two rows of comparators. Due to the spatial continuity of the analog signal, the difference is usually relatively small. Therefore, this quantization method can reduce the number of flips of the D flip-flop, thereby reducing the power consumption of the count. In addition, in traditional TDC-based SS ADCs, an edge detection circuit is usually needed to generate a stop signal, that is, the comparator flip signal is delayed, and the D flip-flop is used to sample the end of the clock cycle where TDC quantization is located, i.e., the rising edge of the next clock cycle. In the improved TDC, the stop signal of the faster row in the comparator flipping signal is the end of the whole quantization time, and the stop signal of the slower row in the comparator flipping signal is the stop signal of the other row in the TDC. Therefore, the improved TDC does not need the edge detection circuit to provide the stop signal, which can further save power consumption.