Access Official Technical PDF

Share Summary:LinkedInEmail

Silicon photomultipliers (SiPMs) are used in imaging and detection systems where compactness, ruggedness, low operating bias and insensitivity to magnetic fields are valuable. Their small form factor also makes it practical to assemble two-dimensional sensor arrays. The challenge is deciding how to acquire signals from those arrays: a SiPM array readout must balance the number of electronics channels against image quality, noise, timing and the ability to distinguish events.

There is no single architecture that suits every detector. Independent readout preserves the most information from each pixel, while multiplexing reduces the number of amplifiers and data-acquisition channels. The right choice depends on the application’s resolution and timing requirements, array size and tolerance for performance trade-offs.

Independent readout: maximum pixel-level information

In a 1:1 arrangement, each sensor pixel has its own amplification and data-acquisition channel. An array with N pixels in one direction and M in the other therefore requires N × M readout channels. This is the most direct way to preserve the signal from each pixel.

Because channels are acquired independently, this approach can support identification of multiple interactions or scatter events, correction for pixel non-uniformity and individual bias control where needed. It also provides the highest signal-to-noise performance among the approaches described and supports the highest count rates without system pile-up.

The cost is channel count. Independent readout can be practical for small systems using discrete electronics, or for larger systems using an application-specific integrated circuit (ASIC). When neither option suits the system’s complexity or cost constraints, multiplexing offers a way to reduce the electronics burden.

SiPM array readout through multiplexing

Multiplexing combines signals from multiple pixels so that fewer readout channels are needed. This can reduce electronics and data-acquisition complexity, but the combined signals may make it harder to identify simultaneous events, account for scatter or correct pixel-to-pixel variations. Connecting multiple SiPMs can also raise the capacitance seen by a channel, affecting noise and timing.

Several architectures encode position in different ways. Their channel savings should be considered alongside the information the system needs to preserve.

Anger logic and discrete positioning circuits

Anger-style readout uses resistive networks to distribute a pixel’s signal among position-encoding outputs. In a typical arrangement, signals are combined into X+, X−, Y+ and Y− channels. Comparing the paired outputs provides an estimate of the interaction’s position. This method is the basis of the Anger camera and is also used as a position-decoding principle in other multiplexed designs.

A discretized positioning circuit (DPC) uses resistor chains to distribute signals through an array. The position can be calculated from four outputs, reducing the channel count for an M × N array to four. The circuit is passive and uses relatively few resistors per node. However, charge division can be asymmetric between the two axes, distorting the image or weakening pixel definition in one direction. Performance also tends to degrade as pixel count grows, so DPC is generally better suited to smaller arrays.

Row/column and symmetric charge division

In row/column, or crosswire, readout, signals from pixels in each row and column are summed. An N × M array needs N + M channels rather than one channel per pixel. With SiPMs, the pixels can be wired so that columns form one set of outputs and rows form another. Combining the resulting X and Y signals with Anger logic estimates the event location.

The shared connections increase the capacitance and noise at the readout, potentially affecting energy and timing performance. The extent of the impact depends on system-level factors. Symmetric charge division (SCD) extends the row/column approach by adding resistive chains to reduce the outputs to four: two for each axis. Amplifiers can be added before the chains to improve signal-to-noise ratio and reduce the effects of sensor capacitance and input crosstalk, at the cost of extra components, power and complexity.

What detector studies show about the trade-offs

Reported PET detector studies illustrate why channel reduction cannot be assessed in isolation. In one 12 × 12 array study, 144:4 multiplexing resolved 1.5 mm LYSO scintillator pixels at room temperature, although some edge pixels were not clearly resolved. Cooling to 10°C improved pixel resolution in that setup. A less aggressive 144:24 row/column configuration resolved the 1.57 mm pixels to the edges at room temperature and reported energy resolution of 10–11%.

Another study compared resistor values in a DPC circuit. Higher values produced accurate pixel spacing, stronger signal amplitude and better energy resolution, but timing was substantially compromised. Medium resistor values provided a different balance, with reported timing resolution of 3 ns. These results are specific to the tested configurations; they show that resistor selection and operating conditions can change the balance among image formation, energy measurement and timing.

Signal-driven multiplexing for timing

Conventional multiplexing of SiPM fast outputs can combine the capacitance and noise of multiple sensors at a shared node, degrading pulse rise time and height. Signal-driven multiplexing (SDM) uses a pair of fast Schottky diodes at each sensor to isolate the fast outputs. When a sensor receives light, its signal turns on the path to the common output while helping suppress noise from other sensors.

This arrangement is intended to connect the active sensor to the shared timing channel without presenting the same combined load as direct summing. A separate diode bias can further improve response speed and sensitivity, according to the application note. In tests using 3 mm SiPM pixels and Schottky-multiplexed fast outputs, the reported best coincidence resolving time increased from 304 ps for a 16-pixel board to 371 ps for 32 pixels and 473 ps for 64 pixels. These measurements describe the tested boards and conditions, not a universal scaling rule.

Choosing an architecture for the application

For system designers, the practical decision is which information matters most and how much channel complexity the system can support:

  • Choose 1:1 readout when independent pixel information, event separation or timing performance takes priority and the electronics budget allows it.
  • Consider row/column readout when reducing channels is important but retaining separate row and column signals is useful.
  • Evaluate DPC or SCD when a four-channel position-encoded output is attractive, while accounting for image distortion, capacitance and timing effects.
  • Assess SDM when the design objective includes multiplexing fast outputs for timing, and validate diode bias and array-size effects in the intended circuit.

Ultimately, SiPM readout architecture is a system-level trade-off, not simply a way to minimize channel count. Evaluate the required spatial resolution, event rates, energy and timing performance alongside noise, temperature, bias control, power and electronics complexity. Testing the complete detector configuration is essential: results from one array, scintillator arrangement or resistor network should not be assumed to transfer unchanged to another.

Access Official Technical PDF

About The Author