1. Introduction
Today’s smartphone displays are becoming slimmer due to the form factor of these devices. In these slim display modules, the display and touch screen panel (TSP) are placed extremely close to each other, thereby increasing the coupling capacitance between the electrodes of the two panels. The increased capacitance makes it easier for the display drive signals to interfere with the sensing signals of the TSP, which can significantly degrade touch performance.
To address the display noise problem, a number of circuit techniques have been proposed [
1,
2,
3,
4,
5,
6]. One such technique is to subtract the measured noise in a display common plate from the input signals of the touch-sensing ICs [
3]. However, because this method uses the average noise measured in the full display area, it does not reflect local differences in noise depending on the position of the sensor channels. Another method is to separate the touch-sensing period from the display driving period [
7]. Since this method does not generate any display noise during the touch-sensing period, the signal-to-noise ratio (SNR) against the display noise is theoretically infinite. However, in high-resolution displays, the display quality may worsen due to the insufficient writing time of a display image. In addition, it can also reduce the touch-sensing time, which may seriously degrade immunity against noise sources other than the display noise. Increasing the driving voltage for TSPs can also be a useful solution, but it requires a fabrication process that forms transistors with high breakdown voltages and consumes higher power to charge and discharge parasitic capacitance around the sensor electrodes [
8].
In other methods, analog front-end (AFE) circuits using differential sensing schemes have shown excellent noise immunity against display noise by exploiting the fact that display noise coupled to adjacent channels has almost the same magnitude and phase [
2,
3,
4,
5,
6,
9,
10,
11,
12]. Because they calculate touch coordinates by using the difference between the signals sensed in two adjacent channels, the display noise can be significantly reduced. The differential sensing methods for touch sensing have been implemented in various forms to date. The first scheme used the output difference between two single-ended charge amplifiers (SE-CAs), but each output is easily saturated to the supply or ground by strong display noise in spite of the simple implementation [
9,
10,
12]. In a more advanced circuit, pseudo-differential amplifiers in which one input of the amplifier is fixed to a reference voltage were used, but unfortunately, this approach still suffers from the same saturation problem [
2]. To prevent the charge amplifier output from saturating, several techniques that apply fully differential charge amplifiers (FD-CAs) at the first stage of an AFE were proposed [
6,
11]. Among them, two studies, [
6,
11], showed particularly excellent display noise immunity, both of which implemented input common-mode feedback (ICMFB) circuits as well as FD-CAs. The ICMFB circuits strongly force the average voltage of two inputs of the FD-CA to a specific reference voltage, typically half the supply voltage, so that the input voltages do not obviate from the input dynamic range of the amplifier. In this scheme, since the auxiliary ICMFB circuit, not the main amplifier, supplies current to the TSP sensors in response to display noise, the FD-CA outputs can avoid saturation. Nevertheless, because of the small differential feedback gain of the ICMFB circuit, the differential components in the noise generated from the ICMFB circuit can distort touch signals in the FD-CA outputs, resulting in a reduced SNR. This problem will be discussed in detail in
Section 3.
The work presented in this paper also implemented a fully differential amplifier at the first stage of AFE circuits. However, instead of using an ICMFB circuit, by analyzing the magnitude of the CA input terminal voltage fluctuating in response to display noise, we designed an FD-CA with a wide input dynamic range that can cope with the worst display noise. The AFE circuit with this amplifier has two distinct advantages compared with the previous work. First, there is no SNR decrease due to the differential noise of the ICMFB circuit. Second, even in ultrathin touchscreen panels with considerably large coupling capacitance with display electrodes, the AFE can be implemented as a compact and low-power circuit. In ultrathin panels, the ICMFB circuits should have a large current supply capability to charge the large coupling capacitance, resulting in an increase in the power consumption and area of the circuit. However, if the introduced display noise voltage does not exceed the input dynamic range of the amplifier, since the proposed circuit responds to the difference in input capacitances, not the capacitance itself, the performance requirements of the AFE amplifier can be relaxed.
Section 2 analyzes the temporal characteristics of the display noise introduced from the display to the TSP. Then,
Section 3 describes the details of circuit implementation and
Section 4 evaluates the experimental results and performance of the implemented test chip.
2. Transient Characteristics of Display Noise
Although there have been many previous studies on display noise, most of them focused on developing noise-immune circuits. A quantitative analysis of display noise itself has not attracted much attention. In this section, we analyzed the transient characteristics of display noise by establishing an electric circuit model representing displays with a TSP. First, it is necessary to understand the structure of display modules embedded with a TSP and their operation. The TSP is usually located on top of a display panel to increase touch sensitivity, as shown in
Figure 1a, and the common electrode (or cathode electrode) of the display is located on the upper side of the display pixel circuits [
1,
4].
Figure 1b illustrates a display module with numerous data lines and gate lines to write image data to the display pixels. Here, the data lines transfer voltages corresponding to grayscale data to target pixels, and the gate lines turn thin-film transistors (TFTs) on or off to connect a storage capacitor in the pixel to the corresponding data line. Though there are some exceptions [
13], in general, high-resolution displays write an image by the matrix-addressing method used in memory chips [
14]. In such displays, the gate lines are sequentially driven line by line, whereas all of the data lines are driven simultaneously; therefore, most of the display noise results from driving the data lines.
To date, organic light-emitting diode (OLED) displays have been the most popular type of displays in smartphones.
Figure 2a shows the simplified vertical structure of the OLED display, which includes display pixels, display driving circuitry, and parasitic components for display noise analysis. In most low-power OLED displays, the pixel current variation is less than 100 uA for a horizontal write period, and the common electrode resistance is about several ohms. Hence, the potential change of the common electrode is only a few millivolts at most, which is significantly smaller than the potential variation caused by coupling the data lines (this coupling effect will be discussed further in the remainder of this section). Thus, the pixel current of OLED displays can be ignored in the electric model. On the other hand, because each pixel changes its internal voltage only when the corresponding gate line is turned on, the display noise coupled through the storage capacitance of the display is also very small compared to the noise coupled with the data lines. In conclusion, the mechanism of generating display noise can be described in such a manner that the voltage driving of data lines for a write operation is transferred to the common electrode through the coupling capacitance between the electrodes.
Based on the above considerations, the displays can be modeled as the simple equivalent electric circuit shown in
Figure 2b. The write operation of the data lines is modeled with the display noise source V
D(t). R
D is the equivalent parallel resistance of the routing resistors of all of the data lines, C
D is the coupling capacitance between all of the data lines and the display common electrode, and R
C is the resistance from the display common electrode driver to the common electrode. The coupling capacitance between a sensing electrode and the display common electrode is denoted by C
R. Finally, for simplicity of the circuit analysis, the readout circuits of AFEs can be modeled as an impedance of Z ranging from zero to infinity.
The maximum voltage of node V
A can be calculated from the circuit in the gray box of
Figure 2 b, assuming that an impedance magnitude of C
R is sufficiently greater than R
C. This assumption is valid for most mobile OLED displays. The current density of pixels required to represent the maximum grayscale is approximately 0.1 nA/um
2 [
15], and several hundred microamperes of current may flow through the common electrode in a 6-inch display with FHD resolution (1920 x 1080 pixels). Hence, R
C is usually managed as less than 10 Ω to prevent non-uniform display images due to IR (current x resistance) drop in the common electrode. Meanwhile, in recent years, the thickness of encapsulation layer in flexible OLED displays is challenging to even less than 10 um for long-term bending reliability [
16], and the decrease in thickness leads to a significant increase of C
R. Excessively large C
R limits the operating frequencies of the touch sensing signal by lowering the upper cut-off frequency of the touchscreen [
10,
17], so many commercial displays use mesh-shaped metal electrodes to ensure that its capacitance does not exceed 1 nF per channel. In addition, since most touch sensors operate at frequencies of 1 MHz or less, the calculated maximum V
A voltage can be regarded as the maximum peak of the display noise in the time domain.
Equations (1) and (2) are differential equations describing the gray box circuit. V
D(t) can be expressed as a step pulse with a slope to emulate a write operation of data lines so that it consists of two periods, a slewing period and a settling period, as shown in Equation (3). Here, the slewing time, t
1, is determined by the slew rate of the source amplifiers in the display driver ICs, and the height of the pulse, A, can have a maximum value corresponding to the difference between the two grayscale voltages for black and white.
Solution:
- i)
- ii)
- iii)
The mathematical solution of V
C(t) is summarized from (4) to (6), and its graphical shape is shown in
Figure 3. The solution of (6) means that display noise increases as t
1 decreases or when R
D is much smaller than R
C. However, reducing t
1 below τ increases the power consumption of the display driver ICs and does not effectively improve the write speed of the display because the passive network of the data lines determines the slope of the input pulse written to the pixels. Conversely, if t
1 is longer than τ, the display pixels may not be sufficiently charged within a horizontal writing period, and therefore the image quality of the display may deteriorate. Hence, normal displays set t
1 to a value very close to τ. On the other hand, in OLED displays, R
C typically has a smaller value than R
D because a large R
C makes the write time of the display longer, resulting in an insufficient settlement of the image data in the display pixels. Moreover, a large R
C value may cause a non-uniform display image over the entire display area due to a large IR drop, as discussed before. Therefore, R
C/(R
C + R
D) is expected to have a value of about 0.5 or less in most panels.
Applying the conditions considered above for slewing time and resistance values to Equation (6), we can conclude that the temporal maximum value of the display noise is less than 30% of the maximum voltage change of the display driving pulse.
4. Results and Discussion
Figure 12 is a microphotograph of the test chip fabricated using a 0.35-μm 1P4M process. The test chip was composed of TX drivers that can drive 15 TX sensor electrodes and six columns of AFEs and delta-sigma ADCs to process 12 RX sensor electrodes in a TSP. In the experiment, the TSP with 15-TX and 10-RX sensor electrodes was used. Decimation filters for the delta-sigma ADC, registers to control the TSC and a timing controller were also designed in the digital block of the test chip.
To mimic a display module with large coupling capacitance, two thin metal sheets were stacked beneath the TSP and electrically isolated. Here, the distance from the RX electrode to the upper sheet is approximately 150 um. The upper metal sheet was connected to ground through a small resistor to operate like a display common electrode. A pulse that emulates display driving signals was inserted into the lower metal sheet through an RC low-pass filter to generate display noise. By properly selecting the values of the resistor and the capacitor in the filter, the slope of the pulse was adjusted to be similar to the actual waveform.
Figure 13 depicts the experimental environment for evaluation.
Figure 14 shows the waveform measured at the FD-CA output when the (TX1, RX1) sensor node was touched. Here, the output of TX0 was used as a trigger signal to ensure that the measuring probes did not affect touch operation. When a 6-mm-diameter metal pillar touched the screen, the FD-CA output had a larger voltage swing than that of the other nodes. On the other hand, waveforms with small amplitudes appeared at locations where no touch occurred, which may have been caused by a mismatch of mutual capacitances between RX channels. The touch data offsets resulting from the mismatch were eliminated by subtracting the touch data measured during operation time from the premeasured data in the no touch state. The touchscreen data in
Figure 15 demonstrates that the offset cancellation technique is effective to mitigate the mismatch.
The immunity of the proposed FD-CA against display noise is shown in
Figure 16. A driving signal for display used a square wave with a 100-kHz frequency, 1-us rising or falling time, and 5-V swing. The resistance connected to the upper sheet was adjusted so that the excited voltage of the common electrode by display stimulus returns to the reference voltage within one writing period. The waveform of the first row in
Figure 16 demonstrates that even strong display noise with a peak of about 1 V has little effect on the output of the FD-CA.
To evaluate the operation of the two-dimensional touch sensor, the TX driver generated square waves with a voltage swing of 3.3 V at 240 kHz and drove four channels simultaneously. In addition, 30 pulses were assigned to one chip of the orthogonal code, thereby driving at a frame refresh rate of 250 Hz.
Figure 17 shows distribution profiles of the touch data with and without display noise when a 6-mm-diameter metal pillar touched the center of the TSP, resulting in SNRs of 41 dB and 47 dB, respectively. Compared to the case without display noise, the fluctuation of the touch values slightly increased when the noise was applied, resulting in a decrease of the SNR. Nevertheless, if touch coordinates are calculated by applying the weighted average to the touch data in the 3 x 3 nodes around a touch, the SNR of 41 dB means that the maximum accuracy error is about 0.4 mm on a TSP with a 6-mm sensor pitch. However, in most cases, the accuracy error is much smaller than the maximum value because the maximum error can occur only when the noise of 3σ is symmetrical with an opposite sign in the diagonal direction of the touch position, which rarely occurs. Therefore, the above SNR performance confirms that the test chip has excellent noise immunity even in a strong display noise environment.
Table 2 summarizes the performance of the test chip and compares it with the previous remarkable readout ICs for touch sensing. Compared to the study [
11] using the most similar structure, the proposed test chip showed a slightly smaller SNR in the environment without external noise. However, when the noise was strong, the test chip achieved a higher SNR than the other previous work. On the other hand, a recent study [
21] shows outstanding performance in terms of SNR. Unfortunately, the accurate comparison between the two works is difficult because the evaluation conditions such as display noise pattern are not clearly presented in [
21].
In summary, a quantitative analysis of the characteristics of display noise in the time domain is presented in this paper. Based on the analysis, it is demonstrated that the proposed fully differential amplifier with an input dynamic range wider than the maximum display noise peak can effectively suppress display noise with low power consumption and compact size, even in thin displays. The proposed circuit approach to display noise has potential as a useful solution in emerging displays such as flexible OLED displays.