Touch fingerprint sensor based on sensor cell isolation technique with pseudo direct signaling

Figure 1(a) shows the fingerprint sensor circuit proposed in Yeo’s (2016) study. The sensor cell is used not only as a sensor cell but also a signal transmission cell and an isolation cell that performs different functions. The sensor cell evaluates the electric fields emitted from a finger. The transmission (Tx) cell transmits the signal for sensor evaluation. The isolation sensor cell prevents unwanted signal influx emitted from the Tx cell. CMOS switch connected to EN and NMOS connected to ENISOL in sensor cell determine the function of the sensor cell. Figure 1(b) shows the implemented sensor cell layout. The sensor cell dimension is 50 μm × 50 μm, so that the resolution of the implemented fingerprint sensor is 508 pixel-per-inch (ppi).

Figure 1

Figure 2 shows the top view of the 25 fingerprint sensor cell array of the implemented fingerprint sensor. Each sensor pixel size is 50 μm × 50 μm. As an example, the center sensor cell is for evaluation, and the adjacent eight sensor cells are isolation cell for preventing unwanted electric fields emitted directly from the Tx cells as if building an isolation wall. The implemented fingerprint sensor can change the wall of isolation cells variably. The wider the wall, it clearly blocks the unwanted signal emitted from the Tx cells.

Figure 2

In general, the fingerprint sensor for the mobile application should be molded to ensure reliability. Thicker the passivation film, it is hard to deliver a signal to a finger through molding or coating film. Therefore, the wall of isolation sensor cells should be required because the small electric fields from Tx cells can affect a lot of influence on sensor cell evaluation.

A 96 × 96 array fingerprint sensor was implemented and fabricated with 0.25 μm standard CMOS technology. A fingerprint sensor of 96 × 96 sensor cell array with pseudo-direct signaling scheme, as described in the previous section, was implemented. And the pipelined scan architecture was also applied to the fingerprint sensor (Yeo, 2016). Figure 3 shows the implemented 96 × 96 array fingerprint sensor with a pipelined scan driver architecture. The fingerprint sensor is consisted of 96 × 96 sensor cells array with YDEC and XMUX, eight shift ring counter. The eight shift ring counters generate pipelined scan signals which provide parallel scan of the maximum eight sensor cells simultaneously. Two-stage of 8 by 12 analog XMUX selects a proper column output for the proper evaluation sensor cell. An 8-bit ADC, converts the sensing signal output with 256 gray level, provides the superior fingerprint image quality with 508 ppi resolution.

Figure 3

The sensed signal is fed into an 8-bit ADC through an analog XMUX. The ADC data can be serialized by SPI protocol module or stored in memory, which are not shown in Figure 3. The converted serial data by SPI are interfaced with ARM system and PC for the performance evaluation.

The eight column signal generators, V_CK1~8, generate a reset and an evaluation signals for the column sensor cells. Every eight column cell receives the same clock signal generated from the column signal generator. Therefore, the depth of the pipelined scan can be extended up to 8 which means the eight sensor cells can be evaluated simultaneously.

The eight shift ring counter chooses column clock generators at every 16-clock interval. It is designed that the evaluation of the sensor cell is performed every 16 clocks. Therefore, the pipelined evaluation of the fingerprint sensor cell can reduce the fingerprint image capture time effectively without SNR degradation (Yeo, 2016).

Figure 4 shows the example of the pipelined scan fingerprint sensor driver’s simple timing diagrams (Yeo, 2015a). This example shows the three pipelined scan mode in which the three sensor cells are evaluated at the same time. V_CK1 generates a reset signal for the first sensor cell and an evaluation starts. After the evaluation of first sensor cell has started, V_CK2 generates a reset signal for the second sensor cell and an evaluation of the second sensor cell proceeds simultaneously. V_Ck3 works in the same way. Because there are eight V_CKs in the logic, the ninth sensor cell receives a reset signal generated from V_CK1. Consequently, the evaluation of the sensor cells is performed at every 16 clock after the initial latency, which depends on the depth of the pipelined scan. As shown in Figure 4, the sensing signal is integrated, so the output voltage of the sensor cell linearly increases.

Figure 4

The output voltage of the one-time integration is determined by the capacitance formed between the sensor plate and the fingerprint as expressed in the following equation (Yeo, 2015b): (1) Δ V out ( t ) ≅ − C finger C finger + C S V dd where C_finger is either C_ridge or C_valley in Figure 1.

The voltage difference between the output voltages of a ridge and a valley can be expressed in the following equation (Yeo, 2015): (2) V diff = ( C ridge C ridge + C S − C valley C valley + C S ) V dd

Obviously, the sensing output voltage after n-time signal integrations results in the following equation (Yeo, 2015b): (3) V sense = n ⋅ V dif f

Figure 5 shows the full chip layout of the implemented 96 × 96 fingerprint sensor. The dimension of the full chip is 5,000 μm × 6,000 μm, which is mostly occupied by 96 × 96 sensor array. The sensor array is considered as an active sensing area and its size is 4,800 μm × 4,800 μm. YDEC with sensor cell driver logic and an analog XMUX are placed around chip. An ADC, clock generator logics, and memories are placed at the bottom of the chip.

Figure 5

Figure 9(a, b) shows the obtained 96 × 96 fingerprint images from the fingerprint sensor chip with 80 μm-thick epoxy mold and its histogram equalized image of the original image. The histogram equalization uses dynamic threshold for binarization of the image to enhance the fingerprint image so that it can increase the contrast of the original image.

Figure 9

Wiener filter based on local statistics estimated from a local neighborhood of each pixel, which commonly used filtering technique to reduce noise (Greenberg et al., 2002). Figure 10(a, b) shows the fingerprint Wiener filtered fingerprint image with local neighborhood size of 3 × 3, 5 × 5, respectively.

Figure 10

Figure 11 shows the weights of 2D 3 × 3, 5 × 5 Gaussian filter. The 2D Gaussian filter can be simplified with weighted FIR filter. The weights are roughly calculated as shown in Figure 9(a, b).

Figure 11

Figure 12(a, b) shows the fingerprint 2D Gaussian filtered fingerprint image. The weights of the local neighborhood size of 3 × 3, 5 × 5 are shown in Figure 10.

Figure 12

Figure 13 shows the calculation method using trigonometric function to determine the weights of the FIR filter for the 50 µm thick coated film. In this paper, it is assumed that the electric fields from the sensor electrodes to the finger are roughly proportional to only the distance between the sensor electrode and fingerprint. As shown in Figure 12, the distance between the ridges directly above the sensor electrode is 50 µm while the distance to the adjacent ridge can be calculated as roughly 50 µm/cosα.

Figure 13

Figure 14(a, b) shows the calculated weights of the 2D FIR filter for the touch fingerprint sensor with 50 µm-thick coated film. The local size are 3 × 3, 5 × 5, respectively.

Figure 14

Figure 15(a, b) shows 2D FIR filtered fingerprint images. The weights of the local neighborhood size of 3 × 3, 5 × 5 are shown in Figure 13.

Figure 15

The performances of the fingerprint sensor were tested with the images obtained from the implemented fingerprint sensor. The implemented fingerprint sensor chip operates at 1.8 V~2.5 V and its operating frequency is about 10 MHz. The power consumption in fingerprint image capture operation condition is less than 3 mA and the image capture time is less than 20 ms with the help of the pipelined scanning fingerprint sensor cells technique.

The matching performances of the fingerprint sensor, such as false acceptance ratio (FAR) and false rejection ratio (FRR), were evaluated by using Precise Biometrics’ fingerprint software (https://precisebiometrics.com/wp-content/uploads/2014/11/White-Paper-Understanding-Biometric-Performance-Evaluation.pdf). The software provides an optimal solution for fingerprint matching in mobile systems. FAR indicates a measure of the likelihood that a fingerprint sensor will incorrectly approve an access attempt by unauthorized users. FAR is typically expressed as the ratio of the number of false acceptances divided by the number of authentication attempts. FRR indicates a measure of the likelihood that a fingerprint sensor will incorrectly reject an access attempt by an authorized user. FRR typically is stated as the ratio of the number of false recognitions divided by the number of identification attempts. In general, FAR and FRR are related to each other and are inversely proportional to each other. Therefore, only lower FAR or lower FRR might be misleading about the performance of the fingerprint sensor.

In total, 250 fingerprints have been captured and each fingerprint image was captured in three different directions, 0 degree, 45 degree, and 90 degree. The acquired fingerprint images were stored to build an information database. In the matching process, the captured images were passed to the software and a matching analysis was performed by comparing with the established database. Figure 16 shows the FAR versus FRR results of the fingerprint sensor. FRR is about 3% at FAR = 10⁻⁵ and FRR is about 2% at FAR=10⁻⁴, which are an appropriate performance that can be applied to the mobile systems.

Figure 16

Table 1 summarizes the performances of the implemented fingerprint sensor and the comparisons with other fingerprint sensors. The array size of the other fingerprint sensors is more than three times larger than the implemented fingerprint sensor. And they did not test in molded state. The performances of the implemented fingerprint sensor in this work show that it is suitable for mobile applications.