Overall architecture

Figure 1(b) shows a structural diagram of Topmetal-M2. The total size of Topmetal-M2 is $19\text{ mm}\times 24\text{ mm}$, and the active region is approximately $18\text{mm}\times 23\text{mm}$. The pixel array contains pixels of 400 rows × 512 columns with a pixel pitch of 45 μm×45 μm, and it is divided into 16 subarrays with 400 × 32 pixels per subarray. For each subarray, all the pixels share two analog output channels: an energy information channel (CSA channel in Fig. 2) and a time-of-arrival information channel (TAC channel in Fig. 2). The analog signals of the pixels can be driven out by digital time-division multiplexing readout circuitry. Topmetal-M2 can be configured in two distinct modes. In the operating mode (scanning mode), the readout circuitry controls the switching of the pixel row and column switches using a rolling shutter architecture. This enables the synchronous readout of analog signals from the 16 subarrays. Conversely, in the test mode (single-pixel mode), the readout circuitry enables the interruption of row and column switching, thereby establishing a direct connection between a specific pixel and the output buffer. The maximum clock frequency at which the readout circuitry operates is 40 MHz, which corresponds to a maximum frame rate of 3.125 kHz. The peripheral circuitry is distributed along the left and bottom edges of the chip, whereas the IO pads are positioned at the bottom. With an operating voltage of 1.5 V, the overall power consumption of Topmetal-M2 could be reduced to approximately 225 mW, thereby achieving a low power consumption of ~49mW/cm². Compared with Topmetal-M, Topmetal-M2 achieved a reduction in power consumption by approximately 90%. However, this was achieved at the cost of a reduced input range of approximately 79% and a decrease in the charge-voltage conversion gain by approximately 24%. Nevertheless, as long as Topmetal-M2 can satisfy the requirements of scientific experiments, this tradeoff is meaningful. Therefore, Topmetal-M2 is particularly suitable for low-power, low-noise, and large-detection-area applications, such as space science experiments.

Fig. 2Full-screen View

Structure of a pixel unit cell in Topmetal-M2

Pixel unit cell

Figure 2 shows the structure of a pixel unit cell in Topmetal-M2. This unit consists of an n-well/p-well diode sensor, a Topmetal sensor, a CSA, a comparator, a peak-holding circuit, a TAC, dual two-stage source followers, two row selection switches (ROW_SEL), two column-level column selection switches (COL_SEL), and two subarray-level analog output buffers. The layout of an individual pixel in Topmetal-M2 is depicted in Fig. 3. Each pixel is equipped with two distinct charge sensors to collect charges through two distinct mechanisms. The diode is composed of a regular octagonal n-well with a diameter of 3 μm, which has a 4 μm spacing to the surrounding square p-well. It is located in the upper-left corner of the pixel and is connected to the input of the CSA via a capacitor and a switch MAPS_EN. If the MAPS_EN switch is in the off state, the signal from the diode cannot be transmitted to the CSA, facilitating the selection of different charge sensors for various experimental purposes. The ionizing particles cross the depletion region and generate electron–hole pairs. The charges move in the depletion region under the action of an electric field and produce signals. The Topmetal sensor, which is a 14.71 μm×40.50 μm electrode in the top metal layer, is positioned on the left side of the pixel and is directly connected to the input of the CSA. In the center of Topmetal, there is an exposed noninsulated area (green plate in Fig. 2) of 10.71 μm×36.50 μm to collect charges directly from the surrounding space. Each Topmetal is encompassed by a guard ring electrode (covered by a passivation layer) located on the same metal layer with a spacing of 1.5 μm. The width of the guard ring electrode is 1.5 μm. All the guard ring electrodes in the pixel array are connected to the same guard ring pad. The Topmetal and guard ring electrodes form a coupling capacitor C_gt of ~9.8fF. Using the guard ring pad, a charge $Q_{\text{in}}=C_{\text{gt}}\cdot \text{Δ}{V}_{\text{in}}$ can be injected into the Topmetal through C_gt to evaluate the performance of the pixel sensor, where $\text{Δ}{V}_{\text{in}}$ represents the voltage amplitude of the negative-step pulse applied to the guard ring pad. In the scanning mode, the guard ring pad is connected to the ground. When using Topmetal sensor as a charge collection sensor, Topmetal-M2 must be used together with an ionization medium, which typically consists of a sensitive volume of gas or liquid. The charges generated in the ionization zone above Topmetal-M2 drift toward the chip surface and are directly collected by the Topmetal sensor.

Fig. 3Full-screen View

(Color online) (a) Layout of a pixel unit cell in Topmetal-M2. (b) Diode sensor and Topmetal sensor

The charges collected by the pixel sensor are converted into voltage signals by the CSA. The CSA, which consists of a folded cascode operational amplifier, an NMOS transistor M_f, and a feedback capacitor C_f, is one of the most crucial parts of the pixel. Figure 4 illustrates the structure of the amplifier in the CSA. Compared with the telescopic cascode architecture in the first-generation chip Topmetal-M, the folded cascode architecture has a larger common-mode input voltage swing and a larger output voltage swing, which is beneficial for our low-supply-voltage designs. The capacitor C_f is formed by the parasitic capacitance between the two metal layers, and the value is ~1fF. The charge-voltage conversion gain of the CSA is $\text{Δ}{V}_{\text{out}}/Q_{\text{in}}=-1/C_{\text{f}}$, where Q_in is the input charge, and $\text{Δ}{V}_{\text{out}}$ is the output voltage. We use Mf as an adjustable resistor to discharge the charge accumulated in C_f. The decay time constant of the CSA output signal is $\tau =R_{\text{f}}\cdot C_{\text{f}}$, where R_f denotes the equivalent resistance of M_f. By modifying the gate voltage of M_f, we can adjust the value of R_f to achieve the desired decay time of the CSA output signal.

Fig. 4Full-screen View

Structure of the operational amplifier in the CSA

The CSA output is divided into two channels: CSA and TAC. In the CSA channel, the CSA output is connected to two-stage source followers. The first-stage source follower isolates the CSA from the switches, whereas the second-stage source follower enhances the driving capability of the output. Each column shares a common COL_SEL connected to all the ROW_SELs within the same column. Similarly, all the COL_SELs within the same subarray are connected to an analog output buffer. ROW_SELs and COL_SELs are controlled by the readout circuitry. In the TAC channel, the CSA output is connected to a comparator via a coupling capacitor. Assuming that the LATCHB switch turns on at time $T_{\text{on}}$ and the pixel sensor receives the charges generated by the incident particle at time $T_{\text{arrival}}$ (arrival time of the particle), the output of the comparator transitions to a high state after a short delay time $T_{\text{delay}}$. The peak-holding circuit stores the output state of the comparator and initiates the charging of the capacitor in the TAC, resulting in an increasing output voltage with a fixed slope. The size of the charging current is controlled by the bias voltage of the current source in the TAC. Subsequently, the LATCHB switch turns off at time $T_{\text{off}}$, and the charging time $T_{\text{charge}}$ of the capacitor can be determined from the output voltage of the TAC. The arrival time $T_{\text{arrival}}$ can be calculated using the equation $T_{\text{arrival}}=T_{\text{off}}-T_{\text{charge}}-T_{\text{delay}}$, where $T_{\text{off}}$, $T_{\text{charge}}$, $T_{\text{delay}}$ are known (see more details in [25]). When the TAC output is read out, the output of the peak-holding circuit and the capacitor in the TAC are reset by the switch TAC_RESET.

CSA channel

To evaluate the basic performance of Topmetal-M2, comprehensive chip tests were conducted under ambient air and room temperature conditions. As the readout circuitry employs a rolling shutter architecture, an excessively high clock frequency (rapid pixel switching) can have a detrimental effect on the output signal quality. To ensure the optimal signal quality, the clock frequency of the chip readout circuitry was set to 20 MHz (corresponding to 640 μs per frame). Furthermore, a signal generator was utilized to apply a square-wave pulse with a peak-to-peak amplitude of $\text{Δ}V$ to the guard ring pad of the chip. To study the performance of individual pixels, the chip was configured to operate in the single-pixel mode, enabling us to focus on a fixed pixel. As shown in Fig. 5, the CSA_VRST voltage was set to $V_{\text{CSA\_VRST}}=750\text{mV}$, and the square-wave pulse amplitude was set to $\text{Δ}V=50\text{mV}$. The CSA output is a pulse signal with a rise time of ~30 μs and a decay time of ~3.5 ms. Moreover, Fig. 6 illustrates the correlation between the CSA_VRST voltage and decay time for five different pixels. By adjusting the value of the CSA_VRST voltage, the decay time of the CSA output signal can be varied within a range of approximately 0.2-30 ms. The observed discrepancies in decay time among the pixels can be ascribed to two primary factors: the influence of the parasitic resistance on the CSA_VRST voltage of each pixel and the device mismatch in the CMOS process. Figure 7 shows the I/O dynamic range of the CSA for five different pixels. The charge-voltage conversion gain is ~59.56μV/e^-. Within the input charge range of ~0-3ke^-, the pixels exhibit a favorable linear response to the input pulse signal, resulting in a CSA output voltage range of ~0-180mV. A reduction in the operating voltage resulted in a decrease in the output swing of the amplifier, and the cascaded source followers limited the output swing. We observed discrepancies in the saturation levels among the pixels, which can be attributed to two primary factors. The first is the device mismatch in the CMOS process and the second is the effect of the parasitic resistance on the CSA bias voltage of each pixel.

Fig. 5Full-screen View

(Color online) Input square-wave pulse (in pink) and CSA output pulse signal (in blue) of a pixel. (Note that, owing to the design of the readout electronics, the actual output value of the CSA is half of the value displayed by the oscilloscope. The same situation occurs for the TAC output in both Fig. 9 and Fig. 11.)

Fig. 6Full-screen View

(Color online) Relationship between CSA_VRST voltage and decay time for five different pixels

Fig. 7Full-screen View

(Color online) Input/output dynamic range for five different pixels

To analyze the noise performance of Topmetal-M2, a square-wave pulse with an amplitude of 50 mV was applied to the guard ring pad. The CSA_VRST voltage was set to 750 mV. The CSA output was sampled using a 12-bit ADC with a dynamic range from -1 V to 1 V. The primary source of noise within a pixel channel is the CSA amplifier. The noise of a pixel channel can be characterized as the root mean square (RMS) noise voltage at the output divided by the gain of the CSA, resulting in the equivalent noise charge (ENC) given by $Q_{\text{in}}\cdot V_{\sigma }/V_{\mu }$, where $V_{\mu }$ represents the mean value of the output pulse signals, and $V_{\sigma }$ represents the standard deviation. Figure 8 shows the noise distribution of all 204,800 pixels, and Gaussian fitting provides an average value of ~5.3 ADC count (equivalent to ~43.45e^-). Note that this noise value encompasses the contribution from the electronic readout system.

Fig. 8Full-screen View

Noise distribution of all 204,800 pixels

TAC channel

In this section, we present the test results of the TAC channel for Topmetal-M2 operating in both the single-pixel and scanning modes. A signal generator was used to apply a square-wave pulse with a peak-to-peak amplitude of 50 mV as the input signal. The CSA_VRST voltage was set to 750 mV. When the LATCHB switch was turned off at time $T_{\text{off}}$, the TAC output voltage gradually decreased instead of remaining constant, as expected, owing to the leakage current of the reset MOS. All the pixels shared a common LATCHB pad, and it was not possible to read the TAC output voltages of all the pixels simultaneously. To solve this problem, the LATCHB switch was maintained continuously on, and the charging of the capacitor was halted using the TAC_RESET switch. Compared with Topmetal-M, Topmetal-M2 offers two TAC reset methods. The first method is a global reset, in which all the pixels are reset simultaneously using TAC_RESET (global). The second method is a local reset, where the reset signal TAC_RESET (local) generated by the readout circuitry resets the previous pixel while reading out the current pixel.

As shown in Fig. 9, the falling edge of the input signal ($T_{\text{arrival}}$) occurs between $T_{\text{on}}$ and $T_{\text{off}}$. The TAC output voltage increases with a constant slope of $K_{\text{charge}}$. When TAC_RESET turns on at time $T_{\text{off}}$, the TAC output voltage transitions from $V_{\text{TAC\_OUT}}$ to the baseline level. A 12-bit ADC with a dynamic range of -1 V to 1 V was used to sample the TAC output voltage. Assuming that the saturation value of the TAC output voltage is $V_{\text{TAC\_OUT\_full}}$, which corresponds to the maximum charging time $T_{\text{charge\_full}}$, the time resolution can be calculated using $1\text{LSB}=T_{\text{charge\_full}}/\left(V_{\text{TAC\_OUT\_full}}/2\cdot 2^{12}\right)$. The time measurement range of the TAC is from 1 LSB to full-scale $T_{\text{charge\_full}}$. The precision of the TAC depends on its linearity and the resolution of the ADC. The temporal resolution of a TAC largely depends on its charging current. Increasing the charging current can enhance the time resolution but reduce the time measurement range. However, excessively high charging currents can easily cause the TAC to saturate. Therefore, the TAC must be read out and reset in time after a signal event occurs. The readout time depends on the time required to scan the chip frame. Therefore, selecting an appropriate charging current that fulfills the test requirements is crucial. As the TAC reset period of a pixel in the scanning mode is 640 μs, it is necessary to ensure that $T_{\text{charge\_full}}\ge 640\mu \text{s}$. First, we conducted a chip test in the single-pixel mode. We maintained $T_{\text{off}}-T_{\text{arrival}}=T_{\text{charge}}+T_{\text{delay}}=650.00\mu \text{s}$ as a constant and obtained $K_{\text{charge}}=460.81\text{V/s}$; thus, the time resolution is $1\text{LSB}=1.06\mu \text{s}$. The distribution of $V_{\text{TAC\_OUT}}$ is shown in Fig. 10, and the time measurement precision is ~6.55 μs. Subsequently, we conducted a chip test in the scanning mode. As shown in Fig. 11, using a square-wave pulse as the input signal, the TAC output of all the pixels responded to this signal. We equated $T_{\text{off}}-T_{\text{arrival}}$ to $200.00\mu \text{s}\mathrm{,400.00}\mu \text{s}\mathrm{,600.00}\mu \text{s}$ and obtained $K_{\text{charge}}=390.20\text{V/s}$; hence, the time resolution is $1\text{LSB}=1.25\mu \text{s}$. Fig. 12 shows the distribution of $V_{\text{TAC\_OUT}}$ with $T_{\text{off}}-T_{\text{arrival}}=600.00\mu \text{s}$; the time measurement precision is ~7.41 μs. Note that all the parameter values were obtained from a test chip. Different chips may have varying parameter values owing to differences in the chip production process and other conditions. Therefore, it was necessary to calibrate each chip individually.

Fig. 9Full-screen View

(Color online) Input square-wave pulse (in yellow), TAC reset signal (in green), and TAC output signal (in blue) of a pixel (Topmetal-M2 in single-pixel mode)

Fig. 10Full-screen View

Time measurement precision of a pixel (Topmetal-M2 in single-pixel mode)

Fig. 11Full-screen View

(Color online) Input square-wave pulse (in yellow), and TAC output signal (in pink) of a subarray channel (Topmetal-M2 in scanning mode)

Fig. 12Full-screen View

Time measurement precision of a pixel (Topmetal-M2 in scanning mode)

Alpha-particle test

In Topmetal-M2, each pixel contains two types of sensors: a diode sensor and a Topmetal sensor. Both sensors were tested and confirmed to function appropriately. This article presents the test results of a Topmetal sensor. An alpha-particle test on Topmetal-M2 was conducted in ambient air at room temperature. Fig. 13(a) shows the platform used for the alpha-particle test. We utilized a ${}^{\text{241}}\text{Am}$ source, which predominantly emits 5.4 MeV alpha particles, to ionize air. As shown in Fig. 13(b), two metal plates were positioned parallel to Topmetal-M2 with a spacing of 20 mm. The upper metal plate served as the cathode, whereas the lower metal plate served as the anode. The electric field strength between the metal plates is approximately 500V/cm (1000V over a 2 cm air gap). The ${}^{\text{241}}\text{Am}$ source was placed on top of the upper metal plate. A square hole matching the size of Topmetal-M2 was present in the middle of the lower metal plate to allow Topmetal-M2 to receive the charges generated by the alpha particles.

Fig. 13Full-screen View

(Color online) (a) Platform for the alpha-particle test. (b) Structure and principle of the test

As alpha particles pass through the air, they ionize the gas and generate a charge. Subsequently, the charges are attracted to the surface of Topmetal-M2 under the action of an electric field and collected by the Topmetal sensor of each pixel. We can observe the electron clouds generated by the alpha particles imaged using Topmetal-M2. Figure 14 shows two typical tracks of the alpha particles obtained in air. The particle trajectory on the pixel sensor can be observed, with the brightest portion representing the location at which the alpha particles deposited most of their energy. The signal cluster amplitude was determined by summing the amplitude values of all the fired pixels within a signal cluster. Figure 15(a) shows the distribution of the signal cluster amplitude, and Gaussian fitting reveals the mean value of ~860 ADC count. The signal cluster size is defined as the number of fired pixels within a signal cluster. Figure 15(b) shows the distribution of the signal cluster size, with a mean value of ~19 pixels.

Fig. 14Full-screen View

(Color online) (a) (b) Two typical tracks of alpha particles obtained in air

Fig. 15Full-screen View

Result of alpha-particle test. (a) Distribution of signal cluster amplitude. (b) Distribution of signal cluster size

Soft X-ray test

A soft X-ray test was conducted on Topmetal-M2 in a gas chamber at room temperature. Figure 16(a) shows the platform of the X-ray test, where Topmetal-M2 was positioned within a sealed gas chamber and connected to a front-end electronics (FEE) board via pin headers. The chamber was filled with a mixture of helium and dimethyl ether (DME) in a ratio of 30:70 to serve as the working gas. As depicted in Fig. 16(b), the chamber space can be approximately divided into three regions: electron drift, electron multiplication, and charge collection. A gas microchannel plate (GMCP) was employed as the electron multiplier in the electron multiplication region. The GMCP has a thickness of 300 μm, a microhole diameter of 50 μm, and a hole spacing of 60 μm. The distance between the cathode and the upper surface of the GMCP was 10 mm, whereas the distance between the lower surface of the GMCP and the chip was 4 mm. The cathode voltage was set to $V_{\text{drift}}=-3600\text{V}$, the upper surface voltage of the GMCP was $V_{\text{top}}=-1840\text{V}$, the lower surface voltage of the GMCP was $V_{\text{bottom}}=-600\text{V}$, and the anode was connected to the ground. Consequently, the electric field strength is 1.76 kV/cm in the electron drift region, 41.3 kV/cm in the electron multiplication region, and 1.5 kV/cm in the charge collection region.

Fig. 16Full-screen View

(Color online) (a) Platform for the soft X-ray test. (b) Structure and principle of the test

An ${}^{\text{55}}\text{Fe}$ source emitting X-rays with an energy of 5.9 keV was positioned above the cathode. Through the photoelectric effect, these incident X-rays interact with the gas molecules within the electron drift region, leading to their conversion into photoelectrons. These electrons subsequently move to the upper surface of the GMCP and enter the channels through the small orifices. Within these channels, a cascade multiplication of the initial electrons occurs, known as electron gas-avalanche multiplication, resulting in the generation of a significant number of secondary electrons. These secondary electrons exit the channels from the lower surface of the GMCP and are collected by Topmetal-M2 in the anode. Figure 17 shows the photoelectron tracks of the X-rays obtained in a gas mixture of He:DME (30:70). The ionization track of the photoelectrons within the gas chamber was measured. Figure 18(a) shows the distribution of the signal cluster amplitude, and Gaussian fitting reveals the mean value of ~9223 ADC count. Figure 18(b) shows the distribution of the signal cluster size with a mean value of ~117 pixels.

Fig. 17Full-screen View

(Color online) (a) (b) Two typical photoelectron tracks of X-rays obtained in a gas mixture of He:DME (30:70)

Fig. 18Full-screen View

Result of the soft X-ray test. (a) Distribution of signal cluster amplitude. (b) Distribution of signal cluster size