Evaluation of Kyoto's Event-Driven X-ray Astronomical SOI Pixel Sensor with a Large Imaging Area

TL;DR

This paper evaluates the performance of a large-area, event-driven SOI pixel sensor for X-ray astronomy, demonstrating successful imaging, spectral measurement, and noise performance in a new sensor prototype.

Contribution

It presents the development and successful testing of XRPIX5b, the largest SOI pixel sensor with event-driven readout for X-ray astronomy.

Findings

Successful X-ray spectra from all pixels

Readout noise of 46 e- (rms) in frame mode

Effective X-ray imaging in event-driven mode

Abstract

We have been developing monolithic active pixel sensors, named ``XRPIX'', based on the silicon-on-insulator (SOI) pixel technology for future X-ray astronomy satellites. XRPIX has the function of event trigger and hit address outputs. This function allows us to read out analog signals only of hit pixels on trigger timing, which is referred to as the event-driven readout mode. Recently, we processed ``XRPIX5b'' with the largest imaging area of 21.9~mm $\times$ 13.8~mm in the XRPIX series. X-ray spectra are successfully obtained from all the pixels, and the readout noise is 46~e$^-$~(rms) in the frame readout mode. The gain variation was measured to be 1.2\%~(FWHM) among the pixels. We successfully obtain the X-ray image in the event-driven readout mode.

Full text

††journal: Nuclear Physics A

1 Introduction

X-ray charge-coupled devices (CCDs) are standard imaging spectrometers widely used for focal plane detectors in modern X-ray astronomy satellites [1, 2, 3, 4, 5, 6, 7]. CCDs have small pixel sizes of $\sim$ 20 $\mu$m $\times$ 20 $\mu$m and good energy resolution of $\sim$ 130 eV in full width at half maximum (FWHM) at 6 keV. However, CCDs suffer from limited time resolution of $\sim$sec for the conventional MOS-CCDs and $\sim$ 10 msec for the pn-CCDs. Therefore, we have been developing “XRPIX” with a high time resolution better than 10 $\mu$sec for future X-ray astronomy satellites [8, 9].

XRPIX is a monolithic active pixel sensor based on the silicon-on-insulator (SOI) pixel technology [10]. XRPIX contains a comparator circuit in each pixel for hit trigger (timing) and two-dimensional hit-pattern (position) outputs. The function allows us readout of pulse heights only from hit pixels on trigger output timing, which is referred to as the event-driven readout mode. Thus, XRPIX offers a high time resolution better than 10 $\mu$sec and a high throughput reaching $>$ 1 kHz in addition to the imaging and spectroscopic capabilities comparable to those of CCDs. Taking advantage of the good time resolution, we can significantly reduce non-X-ray backgrounds especially above 10 keV due to interactions of cosmic rays by introducing an anti-coincidence technique. In our previous studies, we successfully demonstrated the X-ray detection in the event-driven readout mode [11].

Recently, we processed “XRPIX5b” with the largest imaging area in the XRPIX series. In this paper, we report the first evaluation results of XRPIX5b. We describe XRPIX5b in Section 2. In Section 3, we describe the depletion depth and the X-ray spectrum. The conversion gain uniformity is shown in Section 4. We evaluate the event-driven readout in Section 5, and summarize the evaluation results in Section 6.

2 Device description of XRPIX5b

XRPIX5b is the first device in the XRPIX series having a large imaging area usable for a focal plane detector in X-ray astronomy satellites. We used the 0.2 $\mu$m fully depleted SOI CMOS process by Lapis semiconductor Co., Ltd. Fig. 1 shows photographs of XRPIX5b and the readout board. The imaging area, the format and the pixel size are 21.9 mm $\times$ 13.8 mm, 608 $\times$ 384 pixels and 36 $\mu$m $\times$ 36 $\mu$m, respectively. XRPIX5b was designed based on “XRPIX3b” [12], which has a small imaging area of 1 mm $\times$ 1 mm and almost the same pixel circuit as XRPIX5b. Fig. 2 shows a cross-sectional view of XRPIX5b. XRPIX5b consists of the following three layers: a high-resistivity depleted silicon layer for X-ray detection (sensor layer), a CMOS circuit layer for signal processing (circuit layer), and a buried oxide (BOX) layer for insulation between the two layers. A buried p-well (BPW) is implanted around the sense node with a size of 21 $\mu$m $\times$ 21 $\mu$m for efficient signal charge collection and to suppress the back-gate effect [10]. The sensor layer is fabricated on n-type float zone wafer with a thickness of 300 $\mu$m.

3 Spectral performance

In order to estimate the depletion depth, we obtained 57Co X-ray spectra and measured ratios of 6.4 keV and 14.4 keV X-ray counts with several back-bias voltages ($V_{\rm b}$) from 5 V to 200 V. In the measurements, we cooled XRPIX5b to $-60$ ${}^{\circ}\mathrm{C}$ in order to reduce leakage current. We use all the X-ray events classified as single-pixel, double-pixel, triple-pixel, and quad-pixel events. The details of the readout sequence and spectral analyses are presented by Ryu et al. (2011) and Nakashima et al. (2012)[13, 14]. Fig. 3 (a) and (b) shows the ratios obtained with backside illumination (BI) and frontside illumination (FI) as a function of $V_{\rm b}$, indicating that the ratios are almost unchanged at $V_{\rm b}$ $\gtrsim$ 50 V. This suggests that the quantum efficiencies for 6.4 keV and 14.4 keV are saturated at the $V_{\rm b}$ and the sensor layer is therefore fully depleted at $V_{\rm b}$ $\gtrsim$ 50 V. The ratio with FI is $\sim 20$% lower than that with BI at the full depletion. This is because 22% of 6.4 keV X-rays from the frontside is absorbed by the circuit and BOX layers with the thickness of $9.2\mu$m in total. The full depletion voltage can be translated into a resistivity of 9–10 k\Omega$$\cdotcm, which is comparable to the value of the wafer used for the sensor layer.

We examined the spectral performance with BI and FI in the frame readout mode, in which we read out fixed 32 $\times$ 32 pixels in each frame after every 1 ms exposure. We applied $V_{\rm b}=200$ V to the sensor layer in order to reduce charge diffusion and prevent the charge loss in the sensor layer as much as possible. Fig. 4 (a) and (b) shows 57Co spectra obtained with the central 32 $\times$ 32 pixels with BI (a) and FI (b) using the identical device. We use only single-pixel events which are the events when signal charge is collected within one pixel. The readout noise and the chip output gain are $\sim$ 46 e- (rms) and $\sim$ 18.5 $\mu$V/e-, respectively, both with BI and FI. The leakage current is 3 e*-* msec*-1* pixel*-1*. The energy resolutions at 6.4 keV are 634 $\pm$ 10 eV and 524 $\pm$ 5 eV (FWHM) with BI and FI, respectively. The spectral performance with FI is almost comparable to XRPIX3b [12]. However, a low-energy tail degrades the spectral performance in the case of BI. We speculate that the charge collection efficiency is lower at the backside of the sensor. We will examine the details of the electric fields in the sensor layer using a simulation.

4 Gain uniformity

4.1 Uniformity of the whole area of XRPIX5b

We successfully obtained X-ray spectra from all the pixels in the frame readout mode, and examined the gain uniformity of XRPIX5b. Figs. 5 and 6 show a histogram and a map of the gains of the pixels at each location of the sensor. The gains shown here are obtained by fitting 57Co spectra with BI of single-pixel events from 32 $\times$ 32 pixels, which we refer to as a cell hereafter. The data acquisition conditions are the same as those in Section 3. The statistical errors of the gains obtained by fitting spectra are negligible ($<$ 0.1 $\mu$V/e*-*) to the gain variation. Fig. 5 indicates that the cell-to-cell gain variation is at a 1.2%, which is not problematic for most of our applications. However, the gain variation has a systematic tendency. Fig. 6 shows that the cells away from the analog output pad, which is at the bottom-left corner of XRPIX5b, have lower gains. The possible cause would be voltage drops due to trace resistance of the signal lines.

4.2 Uniformity in a cell

We investigated the pixel-to-pixel gain variation in a cell. Fig. 7 shows a histogram of gains of the cell which locates at [column address, row address] = [48, 80]. The other cells have the same tendency as Fig. 7. The gain variation in a cell is $\sim$ 2% (FWHM), which probably reflects the gain variation of the amplifiers in each pixel. The gain variation has little effect on the energy resolution of 8–10% (Fig. 4).

5 Event-driven readout

We evaluate the event-driven readout in XRPIX5b in this section. The details of the readout sequence in the event-driven readout mode are presented in Takeda et al. (2013) [11], but there is one different point that the hit address encoder circuit is newly equipped with XRPIX5b. This circuit allows us to know hit address faster through the parallel digital signal line than through the serial digital signal line used by in previous chips. Fig. 8 shows an X-ray image taken with XRPIX5b in the event-driven readout mode. We irradiated the detector with 22 keV and 25 keV X-rays from 109Cd at room temperature. We operated it with $V_{\rm b}=10$ V in this experiment in order to reduce the leakage current. We confirmed that the trigger circuits in all the pixels operate correctly and X-ray hit addresses are properly output. While the event rate in Fig. 8 is $\sim$ 60 Hz per sensor, we confirmed that XRPIX5b successfully detects X-rays from 241Am at the event rate of $\sim$ 570 Hz per sensor without using the mask, at the conditions of the backside illumination, $V_{\rm b}=10$ V and the operating temperature of $-60$ ${}^{\circ}\mathrm{C}$. Since the event rate was limited by the irradiated X-ray flux, XRPIX5b would tolerate an even higher event rate.

We found a significant problem with X-ray spectrum of XRPIX5b in the event-driven readout mode. Fig. 9 shows 241Am spectra obtained in the frame readout mode (a) and in the event-driven readout mode (b). The data acquisition conditions are the same as those in Section 3. The energy resolution in the event-driven readout mode is much worse than that in the frame readout mode. Takeda et al. (2014) [15] reported that spectral performance in the event-driven readout mode is degraded compared to the frame readout mode due to the capacitive coupling between the BPW directly connected to the sense node and the trigger digital signal line in the circuit layer.

Ohmura et al. (2016) [16] reported that the additional silicon layer acts as an electrostatic shield and successfully reduces the capacitive coupling between the BPW and the trigger digital signal line in the circuit layer. Ohmura et al. (2016) [16] and Miyoshi et al. (2017) [17] applied the Double SOI (DSOI) structure to realize the shield layer. Takeda et al. in prep. [18] will report successful reduction of the capacitive coupling and improvement of the spectral performance in the event-driven readout mode by using the small size XRPIXs with DSOI, XRPIX6D. We are newly processing “XRPIX7” with the DSOI structure and with the same imaging area as that of XRPIX5b. We expect the DSOI structure will drastically improve the spectral performance of XRPIX7 in the event-driven readout mode as compared with that of XRPIX5b.

6 Summary

We processed XRPIX5b, which has the largest imaging area of 21.9 mm $\times$ 13.8 mm, and evaluated its performance. We successfully obtained X-ray spectra from all the pixels. The readout noise and the gain are $\sim$ 46 e- (rms) and $\sim$ 18.5 $\mu$V/e-, respectively, XRPIX5b outputs analog signals with little performance variation among the pixels. The trigger circuits in all the pixels operate correctly even though there is a problem with X-ray spectrum in the event-driven readout mode.

Acknowledgement

We acknowledge the valuable advice and great work by the personnel of LAPIS Semiconductor Co., Ltd. This study was supported by the Japan Society for the Promotion of Science (JSPS) KAKENHI Grant-in-Aid for Scientific Research on Innovative Areas 25109002 (Y.A.), 25109003 (S.K.) and 25109004 (T.G.T. & T.T.), Grant-in-Aid for Scientific Research (B) 23340047 (T.G.T.), Challenging Exploratory Research 26610047 (T.G.T.), Grant-in-Aid for Young Scientists (B) 15K17648 (A.T.) and Grant-in-Aid for JSPS Fellows 15J01842 (H.M.). This study was also supported by the VLSI Design and Education Center (VDEC), the University of Tokyo in collaboration with Cadence Design Systems, Inc., and Mentor Graphics, Inc.

Refference