Real-Time Non-Uniformity Correction without TEC for Microbolometer Array

Abstract

This paper describes a new readout integrated circuit and non-uniformity correction (NUC) method that ensures that the bolometer array has low non-uniformity over a wide operating temperature range without a thermoelectric cooler (TEC). The proposed NUC minimizes the circuit and memory required for signal processing, making it suitable for compact and power-efficient portable infrared cameras. It corrects the aging phenomenon through start-up calibration and corrects non-uniformities without a TEC through calibration during operation mode. It minimizes the calibration process during operation mode and uses a pixel-level analog-to-digital converter to enable real-time NUC. A 0.18 μm standard CMOS process is applied to the proposed NUC. The frame rate for calibration during the operation mode is approximately 14.3 Hz. The proposed NUC demonstrates excellent uniformity with a non-uniformity of less than 0.12% over a wide operating temperature range (−20 to 50 °C).

1. Introduction

Infrared (IR) cameras, which detect objects with infrared radiation instead of visible light, are widely used in various fields [1]. In particular, the demand for compact and power-efficient portable infrared cameras has increased for use in ultralight portable devices, such as drones or head-mounted displays (HMDs). A cooled IR camera lowers the sensor temperature to cryogenic temperatures, which is necessary to reduce thermally induced noise. However, cryogenic cooling systems consume high power and occupy a large volume. Therefore, considerable research has been conducted on microbolometers, which are uncooled infrared sensors that do not require cryogenic cooling systems [2,3,4,5]. The microbolometer is a representative thermal infrared sensor that uses a heat-sensitive material whose resistance changes with temperature. Amorphous silicon (a-Si), thin-film vanadium oxide (VO_x), and titanium oxide (TiO_x) are widely used as heat-sensitive materials.

Since two-dimensional microbolometer arrays have large variations in characteristics between pixels, non-uniformity correction is essential for excellent infrared images [6]. However, the uniformity of a microbolometer array is highly sensitive to changes in operating temperature. Figure 1 shows the variation in resistance with temperature for two bolometers. As the change in bolometer resistance according to the operating temperature is highly nonlinear, the offset and gain differences between the two bolometers vary according to the operating temperature. To solve this problem, a thermoelectric cooler (TEC) can be used to maintain the bolometer array at a constant operating temperature. However, because TECs consume high power and occupy a large volume, they cannot be used for applications in ultralight portable devices [7,8].

Because of the well-established silicon manufacturing methods and the simplicity of its structure, the a-Si-based microbolometer array exhibits uniform characteristics [9]. Therefore, good uniformity can be maintained without a TEC by simply adjusting the offset correction coefficient according to the operating temperature [10]. VO_x and TiO_x microbolometers show superior 1/f noise compared with a-Si microbolometers. However, they show a relatively higher non-uniformity because metal oxides have various phases, and their characteristics change with time due to high bias current. Therefore, good uniformity cannot be obtained only by adjusting the offset correction coefficient for VO_x or TiO_x microbolometers. The adjustment of the gain correction coefficient is additionally required according to the operating temperature [11]. Calibration for each sensor can be performed without a TEC using linear interpolation to adjust the gain and offset correction coefficients according to the current operating temperature [1]. However, there is an additional cost to find a table of correction coefficients for the operating temperature point. Moreover, time-varying characteristics cannot be calibrated.

In this study, we propose a novel readout integrated circuit (ROIC) with a simple structure capable of efficient non-uniformity correction (NUC) over a wide operating temperature range without using a TEC. Offset and gain can be corrected in real time and even the ‘aging’ phenomenon, in which characteristics change over time, can be corrected. In addition, the use of a pixel-level analog-to-digital converter (ADC) enables fast and accurate correction according to the surrounding environment.

2. Basic Concept

2.1. Monolithic IR Camera Chip

Figure 2 shows an outline of the monolithic IR camera chip to which the proposed ROIC can be applied. To reduce the size, weight, cost, and power consumption of the camera, considerable research on wafer-level cameras has been conducted, and microlens and thin-film packaging technologies are needed to realize it [12,13,14,15]. The goal of this study is to implement an ROIC suitable for the application shown in Figure 2, enabling fast and efficient NUC without a TEC. The conventional shutter, which is essential in a general IR camera, can be replaced with a shutter using a liquid crystal (LC). The LC shutter can be used for the NUC of the microbolometer array along with the IR organic light-emitting diode (OLED). IR OLED has a fixed power in the mid-wave infrared (MWIR) band, and it causes a temperature change of about 1 K in the bolometer.

2.2. Calibration Sequence of the Proposed NUC

Figure 3 shows the calibration sequence of the proposed NUC method, which is divided into calibration in the production phase (a), and calibration when power is supplied during use (b). The IR OLED in Figure 2 is the light source for the start-up calibration in Figure 3b, and there is a variation in the amount of light from the IR OLED detected by each sensor. Therefore, a correction coefficient (β_n) to calibrate this variation is necessary and is obtained through the process in Figure 3a. First, as shown in Figure 3a, the operating temperature is adjusted to the standard temperature T₀ using a TEC without any light source, and the bias voltage V_B0n of each sensor is determined by applying a bias current I_B to it. Thereafter, while maintaining the operating temperature T₀ and V_B0n of each sensor, two currents sensed by each sensor are sequentially obtained using the standard light source ΔT₀ and IR OLED, respectively. The standard light is blackbody radiation, and ΔT₀ is approximately 1 K. The correction coefficient β_n (=ΔT_0n/ΔT₀) of each sensor is obtained by comparing the two currents.

The start-up calibration shown in Figure 3b begins when the camera is turned on for use after the production is completed. As this application does not use a TEC, the operating temperature of the sensor is the same as the external temperature T, and the external light is blocked by the LC shutter during the start-up calibration. First, the operating temperature T is sensed, and the bias voltage V_Bn of each sensor and correction coefficient γ are determined. Coefficient γ is a function of T and is necessary for real-time NUC. The process of offset correction by maintaining the bias voltage V_Bn of each sensor if the operating temperature does not change significantly will be explained later. Next, using β_n and the current of each sensor obtained using the IR OLED, a coefficient α_n for gain correction is calculated. When the operating temperature changes significantly during operation, only V_Bn and γ must be updated, enabling real-time NUC. Since the correction coefficients are updated each time the camera is turned on, the aging phenomenon of the sensor can also be corrected. As the IR OLEDs are in use only for a short time, they can maintain constant characteristics regardless of camera usage time.

3. Proposed ROIC for Real-Time NUC without a TEC

3.1. Circuit Operation for Calibration in the Production

Figure 4 and Figure 5 show the schematic of the unit cell circuit and block diagram of the overall arrangement for the proposed ROIC, respectively. As the pixel size (17 μm × 17 μm) of the sensor is very small, it is difficult to implement the unit cell circuit shown in Figure 4 within one pixel. Accordingly, 16 (4 × 4) adjacent bolometers used as IR sensors share one unit cell circuit using 16 switches (ϕ_SR1 − ϕ_SR16) in a time-division manner. The bolometer resistance R_B at temperature T is given by the following equation:

$$R_B\hspace{0.17em}=\hspace{0.17em}{R}_o\hspace{0.17em}\cdot \hspace{0.17em}{e}^{\frac{b}{kT}},$$

(1)

where R_o and b are constants that are determined by the physical properties of the bolometer materials, and k is the Boltzmann constant [16,17]. R_o and b in Equation (1) show a large difference between bolometers.

During the first phase of Figure 3a for determining V_B0n, signals ϕ_s and ϕ_B in Figure 4 are set to the logic ‘1’ state, and amplifier A₁ is used as a comparator. In this case, the connections depicted by dotted lines among those depicted by arrows in Figure 4 are valid. The bias current I_B is supplied to the bolometer of each unit cell using the global current source I_B while the operating temperature is T₀. V_B0n is determined using the following equation:

$$V_{B0n}\hspace{0.17em}=\hspace{0.17em}\hspace{0.17em}{I}_B{R}_o\hspace{0.17em}\cdot \hspace{0.17em}{e}^{\frac{b}{k{T}_0}}.$$

(2)

The V_B0n of each bolometer is input into a column multiplexer and stored in the V_Bn data memory outside the unit cell by a single-slope (SS) analog-to-digital (A/D) conversion process. For the SS A/D conversion process, a rising ramp signal is applied to the positive terminal of A₁ using a column-level bias circuit. A₁ detects when the ramp signal equals V_B0n and enables the digital value of V_B0n from a 12-bit counter to be stored in 12-bit memory.

During the second phase of Figure 3a for determining β_n, signal ϕ_s is set to the logic ‘0’ state, and amplifiers A₁ and A₂ are used for the integrator and comparator, respectively. In this case, connections depicted by red and blue lines among those depicted by arrows are valid. Figure 6 shows the timing diagram of the operation mode after the calibration sequence. The same timing is also applied to the second phase of Figure 3a. Since the 16 bolometers share a unit cell circuit, the timing diagram in Figure 6 is for 1/16-th of a frame. When the frame rate of the operation mode is set to 30 Hz, the maximum time that one bolometer can use the unit cell circuit is 2.08 ms. As certain operations cannot be performed on all unit cells simultaneously, all operations are performed with row-level timing for efficiency.

To maintain the V_B0n of each bolometer, the value stored in the V_Bn data memory is periodically transmitted to the capacitance C_B of each unit cell circuit using the ramp signal generator of the bias circuit. The transmission of V_B0n proceeds row by row, as shown in Figure 6, using signal ϕ_B and the row number, shown in square brackets ([]) in Figure 6. As the circuit used in the process of determining and transmitting V_B0n is the same, an accurate V_B0n can be transmitted without additional errors. First, a standard light source ΔT₀ is applied. The integrator input current I_S0 generated for the bolometer resistance R_B01 is given by Equation (3). When ΔT₀ (≈1 K) is sufficiently small compared to T₀ (≈300 K), Equation (3) can be simply expressed through two approximations as follows:

$$I_{S0}\hspace{0.17em}=\frac{V_{B0n}}{R_{B01}}-I_B^{}=\frac{\hspace{0.17em}{I}_B{R}_o\hspace{0.17em}\cdot \hspace{0.17em}{e}^{\frac{b}{k{T}_0}}}{R_o\hspace{0.17em}\cdot \hspace{0.17em}{e}^{\frac{b}{k(T_0+\Delta T_0)}}}-I_B^{}=I_B\left(\frac{\hspace{0.17em}\hspace{0.17em}{e}^{\frac{b}{k{T}_0}}}{e^{\frac{b}{k(T_0+\Delta T_0)}}}-1\right),$$

(3)

$$\begin{array}{l}{I}_{S0}\cong I_B\left(\frac{\hspace{0.17em}\hspace{0.17em}{e}^{\frac{b}{k{T}_0}}}{e^{\frac{b}{k{T}_0}\cdot (1-\frac{\Delta T_0}{T_0})}}-1\right)=I_B\left(e^{\frac{b}{k{T}_0}\frac{\Delta T_0}{T_0}}-1\right)\\ \hspace{0.17em}\cong \hspace{0.17em}{I}_B\frac{b}{k{T}_0^2}\Delta T_0.\end{array}$$

(4)

I_S0 is integrated into the capacitance C_INT through an integrator composed of A₁ and several MOSs. A buffered direct injection (BDI) input circuit using a feedback amplifier in a common gate configuration has a very low input resistance and stable bias voltage. A current mirror after BDI is used such that the range of the integration voltage can be maximized regardless of the V_B0n value. Since C_INT is reset by using A₂ and ϕ_RST, the effect of the A₂ offset can be eliminated. After integration, an SS A/D conversion is performed using the row-level circuit for the ADC and A₂, and the digitally converted I_S0 is transferred to the outside of the unit cell.

Next, the light source is replaced by an IR OLED ΔT_0n, and the process in Figure 6 is repeated to obtain the digital value of the corresponding input current I_S0n. An approximate expression for I_S0n can be obtained from Equations (3) and (4); accordingly, the correction coefficient β_n is determined by the following equation:

$$I_{S0n}\hspace{0.17em}\cong \hspace{0.17em}{I}_B\frac{b}{k{T}_0^2}\Delta T_{0n},$$

(5)

$${\beta }_n\hspace{0.17em}=\frac{I_{S0n}}{I_{S0}}=\frac{\Delta T_{0n}}{\Delta T_0}.$$

(6)

As seen from Equation (6), β_n is the normalized value of ΔT_0n detected by each sensor, which is stored in a non-volatile memory, such as flash memory.

3.2. Start-Up Calibration

Except for the operating temperature being the external temperature T, the process for determining V_Bn in the start-up calibration of Figure 3b is the same as that for determining V_B0n, as described in Section 3.1. Coefficient γ is required for real-time NUC, and the equations of V_Bn and γ for temperature T are as follows:

$$V_{Bn}\hspace{0.17em}=\hspace{0.17em}\hspace{0.17em}{I}_B{R}_o\hspace{0.17em}\cdot \hspace{0.17em}{e}^{\frac{b}{kT}},\stackrel{}{}\gamma \hspace{0.17em}=\frac{T_0^2}{T_{}^2}.$$

(7)

To determine the coefficient γ, the temperature sensor shown in Figure 5 is used.

Next, ΔT_0n is applied, and the process shown in Figure 6 is performed as described in Section 3.1. An approximate expression for the input current I_S1n can be derived in the same manner as Equation (5), using the gain correction coefficient α_n, which is determined by the following equation:

$$I_{S1n}\hspace{0.17em}\cong \hspace{0.17em}{I}_B\frac{b}{k{T}_{}^2}\Delta T_{0n},$$

(8)

$${\alpha }_n\hspace{0.17em}=\frac{1}{I_{S1n}}\cdot {\beta }_n\cdot \gamma =\frac{k}{I_{B}b}\cdot \frac{T_0^2}{\Delta T_0}.$$

(9)

Coefficient α_n is defined as the reciprocal of the final expression in Equation (4) and includes the gain deviation of each bolometer with respect to the standard light source ΔT₀. In Equation (9), I_B and b include the deviations for each bolometer, whereas the other parameters are independent of the deviations. As the coefficient α_n is maintained only when power is supplied, it is stored in volatile memory, such as SRAM.

3.3. Normal Operation Mode

The LC shutter is kept open and the IR OLED is not used in the normal operation mode. Therefore, the amount of light ΔT_n corresponding to the temperature of the target object is transmitted to each bolometer. Using the bias voltage V_Bn of Equation (7) while the operating temperature does not change at value T, the process shown in Figure 6 is performed as described in Section 3.1. An approximate expression for the input current I_S can be derived in the same manner as Equation (5). Consequently, the final result I_{S_f}, corrected using α_n and γ, is as follows:

$$I_S\hspace{0.17em}\cong \hspace{0.17em}{I}_B\frac{b}{k{T}_{}^2}\Delta T_n,$$

(10)

$$I_{S\_f}\hspace{0.17em}=I_S\times {\alpha }_n\times \frac{1}{\gamma }\hspace{0.17em}=\frac{\Delta T_n}{\Delta T_0}.$$

(11)

As shown in Equation (10), the R_o and offset deviations are eliminated in the expression for I_S, which is obtained while maintaining the V_Bn determined according to the method described above. As shown in Equation (11), the effects of the operating temperature T, b, I_B, and gain deviation are eliminated to obtain a result proportional to the temperature of the target object.

3.4. V_Bn Adjustment and Real-Time NUC

When the operating temperature changes significantly in the operation mode, a large offset current is added to Equation (11). Thus, the dynamic range of the input current is greatly reduced. Therefore, the bias voltage V_Bn should be adjusted if the operating temperature changes by more than 2 K relative to the previous temperature. If the operating temperature changes to T’, V_Bn and γ are adjusted as in the start-up calibration described in Section 3.2, and the following equation is obtained from Equation (7):

$$V_{Bn}\hspace{0.17em}=\hspace{0.17em}\hspace{0.17em}{I}_B{R}_o\hspace{0.17em}\cdot \hspace{0.17em}{e}^{\frac{b}{k{T}^{\prime }}},\stackrel{}{}\gamma \hspace{0.17em}=\frac{T_0^2}{T^{\prime }{}^2}.$$

(12)

The adjustment of α_n is not required, unlike in the start-up calibration. Therefore, it is possible to reduce the calibration time during the operation and enable real-time NUC.

Subsequently, the operation mode proceeds in the same manner as described in Section 3.3. The equations of the bolometer resistance R_B′ for ΔT_n at the operating temperature T′ and I_S obtained from R_B′ are as follows:

$${R_B}^{\prime }\hspace{0.17em}=\hspace{0.17em}{R}_o\hspace{0.17em}\cdot \hspace{0.17em}{e}^{\frac{b}{k(T^{\prime }+\Delta T_n)}},$$

(13)

$$I_S\hspace{0.17em}=\frac{V_{Bn}}{{R_B}^{\prime }}-I_B^{}\cong \hspace{0.17em}{I}_B\frac{b}{k{T}^{\prime }{}^2}\Delta T_n.$$

(14)

Next, the equation for I_{S_f} is corrected using the value of α_n in Equation (9), determined during the start-up correction, as follows:

$$I_{S\_f}\hspace{0.17em}=I_S\times {\alpha }_n\times \frac{1}{\gamma }\hspace{0.17em}=\frac{\Delta T_n}{\Delta T_0}.$$

(15)

As shown in Equation (15), all non-uniformities can be corrected by adjusting only V_Bn and γ as the operating temperature changes.

4. Simulation Results

The proposed ROIC is designed using a 0.18 μm standard CMOS process.

Table 1. Design parameters and characteristics of the target bolometer and the ROIC.

Parameter	Value
Bolometer material	VOx
Resistance	200 kΩ (@ 300 K)
Thermal conductance	0.5 × 10−8 W/K
Emissivity	80%
Temperature coefficient of resistance	2.0%/K
Optics F-number	1
Spectral range	3–8 μm
Pixel size	17 μm × 17 μm
Array size	640 × 512
Frame rate	30 Hz
Power consumption per pixel	<0.61 µW

summarizes the design parameters and characteristics of the target bolometer and the ROIC.

The resistance R_B of the target VO_x bolometer is modeled using Equation (1). Four bolometers are set up with a variation in R_o and b in Equation (1) of 10% each, and their characteristics are shown in the lower part of Figure 7a. If I_B is independent of the operating temperature, the change in V_Bn according to the operating temperature is equal to the change in R_B, as shown in the upper part of Figure 7a. In this case, it is difficult to design a bias circuit and integrator because the range of V_Bn is very wide. As shown in Figure 7b, the range of V_Bn can be significantly reduced by dividing the range of operating temperatures into four sections and using different I_B values for each. I_B values for each section do not need to be constant because the deviation of I_B is corrected, as demonstrated in Section 3.

Figure 8 shows the process of determining V_Bn during start-up calibration. After V_Bn is determined by I_B, a rising ramp signal, V_Ramp, is supplied from the bias circuit. When the V_Ramp value becomes equal to V_Bn, the output voltage V_A1 of comparator A₁ changes, and the digital value of V_Bn is stored in the 12-bit memory of each unit circuit. Considering the speed of comparator A₁, the step time of V_Ramp is set to 1 μs, and the time taken to determine the V_Bn of the entire array is less than 70 ms. Therefore, the frame rate for determining V_Bn is approximately 14.3 Hz, and real-time NUC is possible, as mentioned in Section 3.4. The DC offset effect of comparator A₁ observed in Figure 8 compensates for itself when fed to the corresponding bolometer after V_Bn transmission.

Figure 9 shows the simulated waveforms for verifying the unit cell circuit in the operation mode, which were previously described in Figure 4 and Figure 6. The period for V_Bn transmission is set using signal ϕ_B, and the I_S integration time is determined using signals ϕ_RST and ϕ_H. When ϕ_H changes to logic ‘0′, the integration voltage V(C_INT) remains constant. After the integration of I_S, a 12-bit SS A/D conversion is performed using a row-level circuit for the ADC and comparator A₂. The output voltage V_A2 of A₂ controls the enabled terminal of the 12-bit memory.

Figure 10 shows the mask layout of the proposed unit cell circuit shown in Figure 4. As the pixel size is (17 μm)², the size of the unit cell circuit is {68 (= 17 × 4) μm}². The integrator and comparator in Figure 10 are marked based on the operation mode, and the capacitors occupy most of the layout.

To verify the proposed NUC method, corner simulations are performed using the Cadence Virtuoso. For a conventional calibration method that uses a single correction coefficient table to perform a two-point correction, if the operating temperature changes in the absence of a TEC, the correction table will not fit, and the non-uniformity will increase. Figure 11 shows the non-uniformity of the final digital signal obtained by applying the conventional calibration method to a bolometer with non-uniformity, as shown in Figure 7. The correction coefficient is set according to the operating temperature of 20 °C, and Figure 11 shows that the greater the difference between the operating temperature and 20 °C, the greater the non-uniformity. The offset correction in Figure 11 uses the V_Bn adjustment method, as in the proposed method. In this case, the charge feed-through and leakage current caused by the ϕ_B switch generated during the V_Bn transmission-and-hold process may affect the correction accuracy. Therefore, three types of ϕ_B switches are designed, which are denoted as A, B, and C in Figure 11, indicating an NMOS switch, CMOS switch, and NMOS switch with a dummy switch, respectively.

Figure 12a shows that non-uniformity is greatly decreased by the proposed NUC. A table for β_n is obtained by the calibration in production, as shown in Figure 3a, at an operating temperature of 25 °C, and α_n is obtained by the start-up calibration shown in Figure 3b, at an operating temperature of 20 °C.

Table 2. β_n and α_n for the four bolometers shown in Figure 7.

Ro of RB	βn	αn
315	1.051	3.197 × 107
325	0.9489	3.547 × 107
330	0.9491	3.366 × 107
345	1.050	3.193 × 107

shows the β_n and α_n for the four bolometers shown in Figure 7. Subsequently, the non-uniformity according to the operating temperature is observed by applying the same α_n, regardless of the operating temperature. When using the C switch, the transmission and hold of V_Bn are most accurate, and non-uniformity also decreases relatively. The noise equivalent temperature difference (NETD), which can be used as a measure of non-uniformity, is the smallest possible temperature difference that a thermal imager can resolve. The smaller the NETD value, the higher the image resolution. The most frequently used limiting criteria for NETD are 50 and 100 mK, and the corresponding non-uniformities for these two values are shown in Figure 12a. Figure 12b compares the non-uniformity between the conventional NUC and the proposed NUC and shows the superiority of the proposed NUC.

Table 3. Performance comparison.

	[1]	[10]	[18]	This Work
Pixel pitch	17 μm	17 μm	25 μm	17 μm
Bolometer	VOx	a-Si	VOx	VOx
Power supply	N/A	4 V	9 V	1.8 V
NETD	N/A	<47 mK	<63 mK	<50 mK
Operating temperature [°C]	N/A	10 to 70	−20 to 60	−20 to 50
Aging correction	No	No	No	Yes
Applicability to monolithic IR camera chip	No	Yes	Yes	Yes

shows the performance comparison between the proposed NUC and other methods.

5. Conclusions

We propose a novel NUC capable of accurate and efficient calibration without a TEC that is suitable for compact and power-efficient portable infrared cameras. It corrects not only the basic offset and gain non-uniformity but also the aging phenomenon. The input current is integrated after the bias current is skimmed during offset correction. Thus, high-frequency noise is reduced by using sufficient integration time in a limited pixel area. The basic feasibility of the proposed NUC is verified mathematically. The ROIC and driving method are proposed to efficiently implement the proposed NUC and improve the calibration speed and NETD. The frame rate for the in-use calibration is approximately 14.3 Hz; therefore, real-time NUC is possible. The proposed NUC shows excellent non-uniformity characteristics corresponding to an NETD of less than 50 mK over a wide operating temperature range (−20 to 50 °C).