CMOS Monolithic Active Pixel Sensors (MAPS) found numerous applications in the field of heavy ion physics and particle physics. They are being installed in the STAR Heavy Flavor Tracker \cite{RHIC} and will be used in the Micro Vertex Detector of the future Compressed Baryonic Matter (CBM) experiment \cite{Vertex08}. Moreover, their use is considered for the vertex detectors of the International Linear Collider (ILC) \cite{ILC} and the upgrade of the ALICE-ITS \cite{Musa}. The expected integrated radiation doses in these applications range from several $10^{10}\neqcm$ and few $100~\rm krad$ (ILC) to \mbox{$\gtrsim 10^{13} \neqcm $} and $\gtrsim 1\Mrad$ (CBM).
-MAPS integrate their sensitive volume and the pixel readout electronic on one chip, which is produced with commercially available CMOS processes. As shown in figure \ref{fig:mapssensor}, the sensitive epitaxial layer is surrounded by two layers made from P++ doped silicon. Signal electrons, generated by impinging particles, travel in the epitaxial layer and are reflected back at its interfaces to the surrounding layers until they are collected by a P$_{\rm Epi~ Layer}$/N$_{\rm Well}$-diode. The details of the charge collection process depend significantly on the doping of the epitaxial layer. In case this volume is made from the moderately doped ($\sim 10~\rm \Omega \cdot cm$) silicon as found in standard CMOS processes, the charge collection is dominated by thermal diffusion. Significant improvements can be reached by exploiting the lower doping ($\sim 1~\rm k\Omega \cdot cm$) to CMOS-processes dedicated to commercial optical imaging devices. In this case, the sensitive volume is partially depleted, which accelerates the charge collection and improves substantially the tolerance of MAPS to non-ionizing radiation damage. After this improvement, the tolerance of MAPS to bulk damage was extended to $\gtrsim 10^{14}~\rm n_{eq}/cm^2$ \cite{Mi25,RESMDD2012}. Consequently, their tolerance to non-ionizing radiation dose currently outshine the ionizing radiation tolerance. Therefore, the life-time of MAPS in vertex detectors is limited by the effects of surface damage.\newline
+MAPS integrate their sensitive volume and the pixel readout electronic on one chip, which is produced with commercially available CMOS processes. As shown in figure \ref{fig:mapssensor}, the sensitive epitaxial layer is surrounded by two layers made from P++ doped silicon. N doped implantations in one P++ doped layer form a P$_{\rm Epi~ Layer}$/N$_{\rm Well}$-diode. Signal electrons, generated by impinging particles, travel in the epitaxial layer and are reflected back at its interfaces to the surrounding layers until they are collected by a diode. The details of the charge collection process depend significantly on the doping of the epitaxial layer. In case this volume is made from the moderately doped ($\sim 10~\rm \Omega \cdot cm$) silicon as found in standard CMOS processes, the charge collection is dominated by thermal diffusion. Significant improvements can be reached by exploiting the lower doping ($\sim 1~\rm k\Omega \cdot cm$) to CMOS-processes dedicated to commercial optical imaging devices. In this case, the sensitive volume is partially depleted, which accelerates the charge collection and improves substantially the tolerance of MAPS to non-ionizing radiation damage. After this improvement, the tolerance of MAPS to bulk damage was extended to $\gtrsim 10^{14}~\rm n_{eq}/cm^2$ \cite{RESMDD2012,Mi25}. Consequently, their tolerance to non-ionizing radiation dose currently outshine the ionizing radiation tolerance. Therefore, the life-time of MAPS in vertex detectors is limited by the effects of surface damage.\newline
\begin{figure}
\begin{minipage}{8cm}
\section{Sensor design}
In a first step, some exploratory devices were designed to study the properties of sensing nodes integrated in the novel CMOS process. The chips were named MIMOSA-32, MIMOSA-32ter and MIMOSA-34. Each of those chips host matrices with 32 different pixels types, which vary in terms of pixel pitch, as well as in the sensing node and preamplifier layouts. The 1024 pixels of each pixel matrix are arranged in 16 columns with 64 pixels per column. The columns are readout in parallel and their signal is sent to 16 external ADCs\footnote{Only 8 out of 16 columns were read out due to the limited number of ADCs available in IKF readout system. This restriction remains without impact on the conclusions of our study.}. The readout time is $32~ \rm \upmu s$, which represents the design goal of the future sensors for CBM and ALICE.
-In the pixels discussed in the following, the charge collected by the P$_{\rm Epi~ Layer}$/N$_{\rm Well}$-diode is stored in the parasitic capacitance C of the sensing node and hereafter buffered by means of a source follower (see figure \ref{fig:preamplifier}). The signal charge as well as the accumulated charge generated by the leakage current of the collection diode is cleared by means of a continuous bias, which is realized by means of a permanently opened reset switch based on a NMOS transistor (labeled reset-transistor in figure \ref{fig:preamplifier}). In this particular mode of operation, the switch should act as a high resistivity forward biased diode and the pixel should be equivalent to the self-bias pixels discussed in \cite{Deveaux2010428}.
+In the pixels discussed in the following, the charge collected by the P$_{\rm Epi~ Layer}$/N$_{\rm Well}$-diode is stored in the parasitic capacitance C of the sensing node and hereafter buffered by means of a source follower (see figure \ref{fig:preamplifier}). The signal charge as well as the accumulated charge generated by the leakage current of the collection diode is cleared by means of a continuous bias, which is realized with a permanently opened reset switch based on a NMOS transistor (labeled reset-transistor in figure \ref{fig:preamplifier}). In this particular mode of operation, the switch should act as a high resistivity forward biased diode and the pixel should be equivalent to the self-bias pixels discussed in \cite{Deveaux2010428}.
The charge-to-voltage amplification gain of our CMOS-pixels depends predominantly on the size of the parasitic capacitance of the sensing node. Contributors to this parasitic capacitance are in particular the P$_{\rm Epi~ Layer}$/N$_{\rm Well}$-junction, the drain of the reset transistor and the gate of the source follower transistor. The smaller feature size of the $0.18~\rm \upmu m$-process allows for reducing the size and such the capacitance of those structures, which turns into a sizable potential for improving the amplification gain of the pixel. However, reducing the diode size may come with drawbacks in terms of charge collection efficiency and reducing the size of the transistor gates was reported to cause significant 1/f- and RTS-noise in MAPS used for optical imaging \cite{RTS0.18,RTS0.182}.
According to our measurement standard, we defined the noise as the standard deviation of the dark signal of the individual pixel after performing correlated double sampling, pedestal correction and common mode correction. Details on the related measurement procedure were discussed in \cite{Dev07}. The noise of a pixel matrix is defined as the median of the noise of all individual pixels of this matrix and was measured at a temperature of $T=+20\rm^{\circ}C$. Typically the number of $\lesssim 1\%$ "noisy" pixel could be tolerated. Therefore, in the following, we evaluate not only the median value of the pixel noise distribution but also the noise limit so that $\gtrsim99\%$ of the pixels have a lower noise.
\subsection{Impact of the transistor layout on the noise}
-Figure \ref{fig:Mi18Mi32Mi34Noisevergleich} compares the performances of the \mbox{Pixel A} ($0.18\rm ~\upmu m$, biggest source follower transistor gate) with the one of the established \mbox{Pixel R} \mbox{($0.35\rm ~\upmu m$ feature} size). \mbox{Pixel R} shows a small distribution with an median noise of $10.7 \rm \e$, and $\gtrsim 99\%$ of all pixels indicate a noise below $18 \rm \e$. The noise of \mbox{Pixel A} follows a broad distribution with a median of $19.8\rm \e$ and $\gtrsim 99\%$ of all pixels remain below a noise of $41\rm \e$. Based on the results for Pixel A-C, one may state that reducing the surface of the gate of the source follower transistor increases slightly the gain of the pixel. However, the median noise does also slightly increase, mostly because more pixels with very high individual noise are observed in the presence of a small gate: Once the width of the gate is reduced from $1.5 \mum$ to $0.5 \mum$, the ``99\%-noise'' raises from 41 to 63 electrons.
+Figure \ref{fig:Mi18Mi32Mi34Noisevergleich} compares the performances of the \mbox{Pixel A} ($0.18\rm ~\upmu m$, biggest source follower transistor gate) with the one of the established \mbox{Pixel R} \mbox{($0.35\rm ~\upmu m$ feature} size). \mbox{Pixel R} shows a small distribution with a median noise of $10.7 \rm \e$, and $\gtrsim 99\%$ of all pixels indicate a noise below $18 \rm \e$. The noise of \mbox{Pixel A} follows a broad distribution with a median of $19.8\rm \e$ and $\gtrsim 99\%$ of all pixels remain below a noise of $41\rm \e$. Based on the results for Pixel A-C, one may state that reducing the surface of the gate of the source follower transistor increases slightly the gain of the pixel. However, the median noise does also slightly increase, mostly because more pixels with very high individual noise are observed in the presence of a small gate: Once the width of the gate is reduced from $1.5 \mum$ to $0.5 \mum$, the ``99\%-noise'' raises from 41 to 63 electrons.
-This unintuitive finding can be understood by studying the detailed properties of noise pixels and comparing the results with the observations reported in \cite{RTS0.18,RTS0.182}. As shown in \mbox{figure \ref{fig:Mi32-1-f-noise-CDS-Signal}} and in \mbox{figure \ref{fig:Mi32-1-f-noise-Distribution}}, one observes that the dark signal of a representative noisy pixel varies between three well defined levels. This observation is compatible with the presence a Random Telegraph Signal in the source follower transistor. Random Telegraph Signal is most likely caused by individual defects in the silicon, which may absorb or emit an individual electron. The field of this electron adds to the field applied to the gate of the FET. Therefore, the current passing the FET is modulated to two well separated levels depending on the charge state of the defect. As we apply correlated double sample, which means subtracting the values of two consecutive samples, we observe three levels representing a stable state, the absorption, and the emission of an electron in the defect during the integration time.
+This unintuitive finding can be understood by studying the detailed properties of noise pixels and comparing the results with the observations reported in \cite{RTS0.18,RTS0.182}. As shown in \mbox{figure \ref{fig:Mi32-1-f-noise-CDS-Signal}} and in \mbox{figure \ref{fig:Mi32-1-f-noise-Distribution}}, one observes that the dark signal after CDS of a representative noisy pixel varies between three well defined levels. This signature is the origin of the high noise. The observation is compatible with the presence a Random Telegraph Signal in the source follower transistor. Random Telegraph Signal is most likely caused by individual defects in the silicon, which may absorb or emit an individual electron. The field of this electron adds to the field applied to the gate of the FET. Therefore, the current passing the FET is modulated to two well separated levels depending on the charge state of the defect. As we apply correlated double sample, which means subtracting the values of two consecutive samples, we observe three levels representing a stable state, the absorption, and the emission of an electron in the defect during the integration time.
\begin{figure}
\begin{minipage}[t]{0.49 \textwidth}
Note that, while enlarging the transistor size reduces the RTS, cooling seems not to show a positive impact. This stands in contrast to our observations on RTS-noise originating from the pixel \mbox{diodes \cite{RTS}}.
\subsection{Impact of the transistor layout on the sensor performance}
-The relation between the charge collection efficiency (CCE), the gain and the median noise of the pixels was measured with MIMOSA-34. All pixels were derived from pixel D and the diode size was varied. Figure \ref{fig:Diodesurface} shows the noise and the gain of the different pixels. Moreover, the CCE representing the most probable fraction of charge collected by the seed pixel of a pixel cluster is shown. The CCE was measured by means of a $^{55}$Fe-source.
+The relation between the charge collection efficiency (CCE), the gain and the median noise of the pixels was measured with MIMOSA-34. All pixels were derived from pixel D and the diode size was varied. Figure \ref{fig:Diodesurface} shows the noise and the gain of the different pixels. Moreover, the CCE representing the most probable fraction of charge collected by the seed pixel of a pixel cluster is shown. For the CCE, only the matrices with the same pitch of $XX\mum$ are comparable. The CCE was measured by means of a $^{55}$Fe-source.
One observes that the gain is strongly reduced with increasing diode size and the capacitive noise of the pixels raises accordingly. However, this effect comes with an increase in terms of charge collection efficiency, which raises the signal. This is shown in figure \ref{fig:StoNDiodeSize}, which shows the noise, the most probable signal and the S/N as recorded once the sensor was illuminated by $\upbeta$-rays from a $^{90}$Sr-source. One observes that the S/N, which is defined as the most probable signal in the seed pixel divided by the median of the noise distribution, is in the order of 50 and increases slightly with increasing diode pitch despite the increase of the median noise. Once propagating the ``99\%-noise'' to the S/N, one finds that 99\% of all pixels exceeds 22 and no significant impact of the diode pitch is observed. Note that this S/N is sufficient for reliable MIP-detection and remains fairly above the average S/N of our early successful prototypes like MIMOSA-2.
\section{Radiation tolerance}
-The next step we studied the performances of irradiated sensors. In a first exploratory study, we irradiated some MIMOSA-32 sensors at CERN with X-rays of a dose of up to $10\Mrad$ and tested them hereafter. The dosimetry was carried out by the staff of the facility and is considered to have a $10\%$ absolute precision and the sensors were powered during irradiation. The tests were carried out with a number of different pixel designs showing similar behavior. The preliminary results of one pixel type are shown.
+The next step we studied the performances of irradiated sensors. In a first exploratory study, we irradiated some MIMOSA-32 sensors at CERN with X-rays of a dose of up to $10\Mrad$ and tested them hereafter. The dosimetry was carried out by the staff of the facility and is considered to have a $10\%$ absolute precision and the sensors were powered during irradiation. The sensors were operated at $T=+20\C$. The tests were carried out with a number of different pixel designs showing similar behavior. The preliminary results of one pixel type are shown.
After an irradiation dose of up to $3\Mrad$, we do not observe significant changes in the pixel performance. After a dose of $10\Mrad$, the gain of the sensor dropped by a factor of two (figure \ref{fig:10MradNoiseGain}). The origin of this effect is under investigation. Despite the drop, the sensor remained operational and the median noise increased only moderately from $\sim 20\rm \e$ to $\sim 25\rm \e$.
-Evaluating the signal to noise ratio of the pixels by means of $\upbeta$-rays of a $^{90}$Sr-source, one observes a good median signal to noise ratio of 42, which drops to 33 due to the noise increase (figure \ref{fig:10MradStoN}). Accounting for the width of the distribution, the radiation effect becomes small: 99\% of all pixels show an satisfactory S/N above 20 before and above 17 after irradiation. Again, the noise appears to be dominated by RTS-noise and a solid quantitative understanding of the radiation effects on the noise might require sensors, which are optimized for low RTS. However, the results provide an encouraging first evidence that the manufacturing process and our sensor concept is suited to reach the ambitioned tolerance to some $\rm Mrad$.
+Evaluating the signal to noise ratio of the pixels by means of $\upbeta$-rays of a $^{90}$Sr-source, one observes a good median signal to noise ratio of 42, which drops to 33 due to the noise increase (figure \ref{fig:10MradStoN}). Accounting for the width of the distribution, the radiation effect becomes small: 99\% of all pixels show a satisfactory S/N above 20 before and above 17 after irradiation. Again, the noise appears to be dominated by RTS-noise and a solid quantitative understanding of the radiation effects on the noise might require sensors, which are optimized for low RTS. However, the results provide an encouraging first evidence that the manufacturing process and our sensor concept is suited to reach the ambitioned tolerance to some $\rm Mrad$.
\begin{figure}
\begin{minipage}{0.49\textwidth}
-\includegraphics[width=\textwidth]{10MradStoN.pdf}
-\caption{S/N of the pixel P2 of MIMOSA-32 as function of the radiation dose (see text).}
-\label{fig:10MradStoN}
-\end{minipage}
-\hspace{0.02 \textwidth}
-\begin{minipage}{0.49\textwidth}
\includegraphics[width=\textwidth]{10MradNoiseGain.pdf}
\caption{Noise and gain of the pixel P2 of MIMOSA-32 as function fo the radiation dose.}
\label{fig:10MradNoiseGain}
\end{minipage}
+\hspace{0.02 \textwidth}
+\begin{minipage}{0.49\textwidth}
+\includegraphics[width=\textwidth]{10MradStoN.pdf}
+\caption{S/N of the pixel P2 of MIMOSA-32 as function of the radiation dose (see text).}
+\label{fig:10MradStoN}
+\end{minipage}
\end{figure}
%For older sensors in $0.35\mum$ process, there was a runaway of the transistor working point observed due to built-on charges. The sensors could not be readout after an irradiation dose of $\gtrsim 1\Mrad$. In the $0.18\mum$ process, the $SiO_2$ is thinner and therefore accumulated charge should escape via tunnel effect more easily. This can be confirmed by MIMOSA-32. The reference voltage is much more stable and even a sensor irradiated up to $10\Mrad$ can be operated without any adjustment.
Guided by observations made previously in the field of optical imaging, we studied the relation between sensor capacitance and the RTS - 1/f noise. We find that the use of sensor gates with a length close to the minimum feature size introduces significant RTS-noise into some of the pixels. As the moderate amount of noisy pixels determines the threshold settings on future particle sensor, the advantages of the small gates in terms of reduced capacitance and therefore the improved gain cannot be exploited. Concerning the optimal width of the sensing diode, we find that the increase of noise and of the CCE, which are caused by an increasing diode, do mostly cancel each other out and a very good S/N is reached with diode surfaces scaling from $2 \mum^2$ to $11 \mum^2$. The use of bigger diodes appears slightly preferable.
-Concerning the radiation tolerance, we observe that the devices tolerate a dose of $3\Mrad$ without significant losses in performance while for a dose of $10\Mrad$ a tolerable drop of the gain of the pixels was observed. The origin of this finding is currently under investigation. In any case, the S/N of the device remains satisfactory (above 17 for 99\% of all pixels for the pixel type discussed in this work), which is considered as sufficient for a reliable sensor operation.
+Concerning the radiation tolerance, we observe that the devices tolerate a dose of $3\Mrad$ without significant losses in the signal to noise ratio, measured with $\rm \beta$-rays while for a dose of $10\Mrad$ a tolerable drop of the gain of the pixels was observed. The origin of this finding is currently under investigation. In any case, the S/N of the device remains satisfactory (above 17 for 99\% of all pixels for the pixel type discussed in this work), which is considered as sufficient for a reliable sensor operation.
Over all, we conclude that one cannot straight forwardly exploit the lower feature sizes of the $0.18 \mum$-process studied in order to reduce the capacitive noise. This is as the so far dominating capacitive noise is exceeded by the so far insignificant RTS-noise. However, the source of noise is clearly defined and significant noise improvements should be feasible with moderate modifications of the few relevant transistors. Apart from this complication, the $0.18 \mum$-process studied appears well suited for building CMOS sensors for particle detection. Despite the RTS-noise, most pixel designs show a quite satisfactory S/N. This holds in even after irradiating the sensor to a dose of $10 \Mrad$, which extends the tolerance of MAPS of the MIMOSA-family by one order of magnitude and allows for matching the requirements of the vertex detectors of both, ALICE and CBM in this field.
\bibitem{RHIC}
L. Greiner et al. \emph{Sensor development and readout prototyping for the STAR Pixel detector,JINST 4 P03008} (2009)
-\bibitem{ILC}
- ILD Concept Group,\emph{The ILD Letter of Intent}, (2009)
\bibitem{Vertex08}
M. Deveaux et al.,\emph{Design considerations for the Micro Vertex Detector of the Compressed Baryonic Matter experiment, POS(VERTEX2008)028} (2008)
+\bibitem{ILC}
+ ILD Concept Group,\emph{The ILD Letter of Intent}, (2009) DESY 2009-87
+
\bibitem{Musa}
L. Musa, \emph{Conceptual Design Report for the Upgrade of the ALICE ITS, CERN-LHCC-2012-005. LHCC-G-159} (2012)
+\bibitem{Mi25}
+A.Dorokhov et al. \emph{Improved radiation tolerance of MAPS using a depleted epitaxial layer, NIM-A} 624-2 (432-436) (2010)
+
\bibitem{Deveaux2010428}
M. Deveaux et al.,\emph{Radiation tolerance of CMOS monolithic active pixel sensors with self-biased pixels, NIM-A}
624 2 428-431 (2010) \href{10.1016/j.nima.2010.04.045}
\bibitem{Senyukov}
S. Senyukov et al. \emph{Charged particle detection performances of CMOS pixel sensors produced in a $0.18~\rm \mu m$ process with a high resistivity epitaxial layer, NIM-A} 730 0 115-118 (2013) \href{http://dx.doi.org/10.1016/j.nima.2013.03.017}
-\bibitem{Mi25}
-A.Dorokhov et al. \emph{Improved radiation tolerance of MAPS using a depleted epitaxial layer, NIM-A} 624-2 (432-436) (2010)
-
\end{thebibliography}
\end{document}
jspc29 / radhard.git / commitdiff