Motivation

The field of nuclear physics is currently in a paradigm shift how data acquisition systems are built: Classic systems are triggered, i.e., rely on a hardware decision to define the occurrence of an event, to start the actual data acquisition, that is, the analog to digital conversion. Modern electronics however allow to continuously convert the analog detector signals. This is already exploited in more modern, but still triggered, readout systems which replace analog delays required to accommodate the time taken for the trigger decision with ring-buffer in the digital domain.

The next logical step is to eliminate the hardware trigger altogether, and replace the trigger decision with a data selection realized in software, with the following advantages:

• Since all data is already in the digital domain, latency constraints on the selection algorithm are seriously loosened compared to a hardware trigger.
• The software algorithm can access all detector information, allowing us to better suppress noise and be more efficient.
• A streaming readout is, in principle, less complex. Many problems are moved from hardware to software, where better tooling and more expert knowledge is available and more people can contribute. Other problems are reduced due to less hardware being required, or due to the removal of the event building bottleneck.
• The architecture furthers the convergence of online and offline analysis, leading to better data quality control during data taking and shorter analysis cycles.
• Streaming readout allows the efficient readout of detectors operating on longer timescales like TPCs at high event rates without incurring excessive dead-time, and simplifies the read-out of high-channel count, high-rate detectors as it can be scaled up easily.
• For bandwidth-to-disk-limited experiments (e.g. Run 3 of LHC), a streaming readout combined with online analysis allow to drastically reduce the amount of data stored for each event by pre-processing the raw data to extract features like clusters, or even fully analyze the raw data and store only analysis level data structures. This maximizes the amount of physics that can be extracted from the experiment.

For EIC, the adoption of streaming readout has significant advantages: Current rate predictions indicate that, contrary to LHC Run 3, all raw data can be saved to disk after a first-level zero-suppression, maximizing the physics impact of EIC by allowing for data mining in a completely unbiased data set.

This is the fifth workshop following previous events held at MIT in 2017 (Trigger/Streaming readout) and in Spring 2018 (Streaming Readout II), at Christopher Newport University (CNU) / Jefferson Laboratory in Fall 2018 (Streaming Readout III) and Camogli, Italy (hosted by INFN) (Streaming Readout IV).

The development of streaming readout for the nuclear physics community is driven by dedicated research initiatives, like the “Facility for Innovation in Nuclear Data Readout and Analysis” (INDRA) and the lab-wide Streaming Grand Challenge at Jefferson Lab, the BNL LDRD "High Throughput Advanced Data Acquisition for eRHIC, Particle Physics and Cosmology Experiments" as well as experiments like PHENIX, STAR and sPHENIX (BNL), KM3NeT(INFN), BDX (JLAB) and CBM (FAIR) which all make use of streaming readout.

To facilitate the information transfer between the different groups and to organize EIC related activities, the community has formed the EIC streaming readout consortium, called eRD23 of the EIC detector R&D program.

The workshop is designed to focus on active discussion. We encourage all participants to prepare few slides, to be shown during the discussion. The workshop will cover these topics:

• Data rate requirements / management
• Data format considerations
• Offline/online convergence
• Detector support
• Timing requirements
• Differential Cost/complexity/capabilities
• Test setups / further development