Development of charge reset multiplexing for SiPMs using deep learning architecture

Area of Science:

• Nuclear physics and instrumentation
• Deep learning applications in signal processing
• Detector readout electronics

Background:

• Silicon photomultipliers (SiPMs) are crucial in high-density detector systems.
• Increasing channel counts in SiPM arrays escalates system cost and complexity.
• Efficient readout methods are needed to manage large numbers of detector channels.

Purpose of the Study:

• To introduce a charge-reset multiplexing technique for SiPMs.
• To reduce 16 readout channels to a single output line.
• To maintain per-channel waveform information for advanced signal processing.

Main Methods:

• Each SiPM channel uses a charge-reset preamplifier encoding identity via pulse width.
• Pulses from all channels are summed into a single output signal.
• A deep learning autoencoder reconstructs individual signals from the summed trace.

Main Results:

• Reconstructed signals yielded energy resolutions comparable to non-multiplexed systems (10.84% for 137Cs, 16.12% for 22Na).
• Per-channel signal recovery enabled identification and removal of inter-crystal scatter (ICS).
• Clear flood maps were generated, demonstrating effective ICS rejection.

Conclusions:

• Charge-reset multiplexing significantly reduces channel count without compromising performance.
• The method retains per-channel information essential for advanced processing like ICS rejection.
• This technique is suitable for integrated implementations and broad application in high-density detectors for nuclear medicine.