A Caltech Library Service

A Reconfigurable Neural Network ASIC for Detector Front-End Data Compression at the HL-LHC

Di Guglielmo, Giuseppe and Fahim, Farah and Herwig, Christian and Valentin, Manuel Blanco and Duarte, Javier and Gingu, Cristian and Harris, Philip and Hirschauer, James and Kwok, Martin and Loncar, Vladimir and Luo, Yingyi and Miranda, Llovizna and Ngadiuba, Jennifer and Noonan, Daniel and Ogrenci-Memik, Seda and Pierini, Maurizio and Summers, Sioni and Tran, Nhan (2021) A Reconfigurable Neural Network ASIC for Detector Front-End Data Compression at the HL-LHC. IEEE Transactions on Nuclear Science, 68 (8). pp. 2179-2186. ISSN 0018-9499. doi:10.1109/tns.2021.3087100.

[img] PDF - Accepted Version
See Usage Policy.

[img] PDF - Submitted Version
See Usage Policy.


Use this Persistent URL to link to this item:


Despite advances in the programmable logic capabilities of modern trigger systems, a significant bottleneck remains in the amount of data to be transported from the detector to off-detector logic where trigger decisions are made. We demonstrate that a neural network autoencoder model can be implemented in a radiation tolerant ASIC to perform lossy data compression alleviating the data transmission problem while preserving critical information of the detector energy profile. For our application, we consider the high-granularity calorimeter from the CMS experiment at the CERN Large Hadron Collider. The advantage of the machine learning approach is in the flexibility and configurability of the algorithm. By changing the neural network weights, a unique data compression algorithm can be deployed for each sensor in different detector regions, and changing detector or collider conditions. To meet area, performance, and power constraints, we perform a quantization-aware training to create an optimized neural network hardware implementation. The design is achieved through the use of high-level synthesis tools and the hls4ml framework, and was processed through synthesis and physical layout flows based on a LP CMOS 65nm technology node. The flow anticipates 200 Mrad of ionizing radiation to select gates, and reports a total area of 3.6mm2 and consumes 95mW of power. The simulated energy consumption per inference is 2.4 nJ. This is the first radiation tolerant on-detector ASIC implementation of a neural network that has been designed for particle physics applications.

Item Type:Article
Related URLs:
URLURL TypeDescription Paper
Di Guglielmo, Giuseppe0000-0002-5749-1432
Fahim, Farah0000-0003-1252-1447
Herwig, Christian0000-0002-4280-6382
Duarte, Javier0000-0002-5076-7096
Hirschauer, James0000-0002-8244-0805
Ngadiuba, Jennifer0000-0002-0055-2935
Pierini, Maurizio0000-0003-1939-4268
Tran, Nhan0000-0002-8440-6854
Additional Information:© 2021 IEEE. Manuscript received October 31, 2020; revised February 12, 2021 and April 6, 2021; accepted May 23, 2021. Date of publication June 7, 2021; date of current version August 16, 2021. The work of Farah Fahim, Christian Herwig, Cristian Gingu, James Hirschauer, Llovizna Miranda, and Nhan Tran was supported by the Fermi Research Alliance, LLC through the U.S. Department of Energy (DOE), Office of Science, Office of High Energy Physics under Contract DE-AC02-07CH11359. The work of Javier Duarte was supported by the DOE, Office of Science, Office of High Energy Physics Early Career Research Program under Award DE-SC0021187. The work of Philip Harris was supported by the Massachusetts Institute of Technology University Grant. The work of Vladimir Loncar, Maurizio Pierini, and Sioni Summers was supported by the European Research Council (ERC) through the European Union’s Horizon 2020 Research and Innovation Program under Grant 772369. The authors would like to acknowledge CAD support from Sandeep Garg and Anoop Saha from Mentor Graphics for Catapult high-level synthesis (HLS) and Bruce Cauble and Brent Carlson from Cadence for Innovus and Incisive. They also like to thank the Fermilab application-specific integrated circuit (ASIC) group for incorporating the autoencoder block into the ECON-T ASIC; CMS high-granularity endcap calorimeter (HGCAL) and Jean-Baptiste Sauvan for providing simulated module images for training; and Andre Davide for extensive input on network optimization. They acknowledge the Fast Machine Learning Collective as an open community of multidomain experts and collaborators. This community was important for the development of this project.
Funding AgencyGrant Number
Department of Energy (DOE)DE-AC02-07CH11359
Department of Energy (DOE)DE-SC0021187
Massachusetts Institute of Technology (MIT)UNSPECIFIED
European Research Council (ERC)772369
Subject Keywords:Application-specific integrated circuit (ASIC), artificial intelligence (AI), autoencoder, hardware accelerator, high-level synthesis (HLS), Large Hadron Collider (LHC), machine learning (ML), single-event effect (SEE) mitigation
Issue or Number:8
Record Number:CaltechAUTHORS:20210623-154212057
Persistent URL:
Official Citation:G. D. Guglielmo et al., "A Reconfigurable Neural Network ASIC for Detector Front-End Data Compression at the HL-LHC," in IEEE Transactions on Nuclear Science, vol. 68, no. 8, pp. 2179-2186, Aug. 2021, doi: 10.1109/TNS.2021.3087100
Usage Policy:No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:109550
Deposited By: Tony Diaz
Deposited On:23 Jun 2021 18:26
Last Modified:18 Aug 2021 18:32

Repository Staff Only: item control page