Measurements Measurements and and Simulations Simulations of of - - PowerPoint PPT Presentation

Measurements Measurements and and Simulations Simulations of

• f

Single-Event Single-Event Ups Upsets ts in in a 28-nm 28-nm FPGA FPGA

Consequences Consequences fo for front-end front-end electronics electronics in in PAND ANDA@F A@FAIR AIR

• M. Preston, P.-E. Tegn´

er

(Stockholm University)

• H. Cal´

en, T. Johansson, K. Mak`

• nyi, P. Marciniewski

(Uppsala University)

• M. Kavatsyuk, P. Schakel

(University of Groningen)

TWEPP 2017, Santa Cruz, USA

The FAIR facility (Darmstadt, Germany)

Figure courtesy of FAIR website. Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 2/21 Markus Preston - Stockholm University

The FAIR facility (Darmstadt, Germany)

Antiprotons (p¯

p = 1.5 - 15 GeV/c)

+ Fixed hydrogen target (cluster/pellet) ⇒ p¯ p collision L (start-up) = 1031 cm−2 s−1 First ¯ p beam 2025

Figure courtesy of FAIR website. Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 2/21 Markus Preston - Stockholm University

The PANDA experiment

Figure courtesy of PANDA collaboration website.

¯ p p

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 3/21 Markus Preston - Stockholm University

The PANDA experiment

Figure courtesy of PANDA collaboration website.

¯ p p

Electromagnetic Calorimeter (EMC) ∼ 15000 PbWO4 scintillating crystals (cooled to -25 ◦C) Three parts: Barrel Forward Endcap Backward Endcap 99% of 4π around Interaction Point Photo detectors: APDs/VPTTs Dynamic range: 10 MeV - 14.6 GeV

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 3/21 Markus Preston - Stockholm University

Front-end electronics in the Electromagnetic Calorimeter

Figure courtesy of P. Marciniewski.

◮ FPGA-based sADC ◮ Input: Up to 32 photo

detectors per board

◮ 2 28-nm FPGAs per board

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 4/21 Markus Preston - Stockholm University

Front-end electronics in the Electromagnetic Calorimeter

Figure courtesy of M. Albrecht.

1.3 m

◮ FPGA-based sADC ◮ Input: Up to 32 photo

detectors per board

◮ 2 28-nm FPGAs per board ◮ ∼ 600 boards in whole EMC. ◮ Placed around EMC

• perimeter. In Forward

Endcap, ∼1.3 metres from beam pipe.

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 4/21 Markus Preston - Stockholm University

Front-end electronics in the Electromagnetic Calorimeter

Figure courtesy of M. Albrecht.

1.3 m

◮ FPGA-based sADC ◮ Input: Up to 32 photo

detectors per board

◮ 2 28-nm FPGAs per board ◮ ∼ 600 boards in whole EMC. ◮ Placed around EMC

• perimeter. In Forward

Endcap, ∼1.3 metres from beam pipe.

◮ Exposure to a high flux of

particles ⇒ Radiation effects.

◮ Test using neutron and

proton irradiations.

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 4/21 Markus Preston - Stockholm University

Neutron and proton irradiations

Neutron irradiation

◮ At TSL, Uppsala, Sweden. ◮ Beam with continuous

energy spectrum, 0-180 MeV. Proton irradiation

◮ At KVI-CART, Groningen,

the Netherlands.

◮ Three proton energies: 80,

100, 184 MeV.

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 5/21 Markus Preston - Stockholm University

Neutron and proton irradiations

Neutron irradiation

◮ At TSL, Uppsala, Sweden. ◮ Beam with continuous

energy spectrum, 0-180 MeV. Proton irradiation

◮ At KVI-CART, Groningen,

the Netherlands.

◮ Three proton energies: 80,

100, 184 MeV. General setup:

◮ Board perpendicular to

beam.

◮ Beam covering ∼ half of

board.

◮ One FPGA continuously

monitored and read out.

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 5/21 Markus Preston - Stockholm University

Single-Event Upsets (SEUs) in FPGA configuration

◮ SEU: Radiation-induced transient effect, which may affect

multiple parts of front-end board.

◮ Here: study the effect in the FPGA configuration memory.

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 6/21 Markus Preston - Stockholm University

Single-Event Upsets (SEUs) in FPGA configuration

◮ SEU: Radiation-induced transient effect, which may affect

multiple parts of front-end board.

◮ Here: study the effect in the FPGA configuration memory. ◮ Use IP core for error monitoring and correction

◮ Monitored during irradiations ◮ Can correct Single-Bit Upsets ◮ Can correct Multi-Bit Upsets in adjacent memory frames

(using Error Correction Code)

◮ Cannot correct Multi-Bit Upsets within the same memory

frame

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 6/21 Markus Preston - Stockholm University

FPGA configuration memory tests

Measurement process

Start Configure FPGA Beam on Detect SEU Beam off Correctable? Log file SEU time + address Yes No

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 7/21 Markus Preston - Stockholm University

SEU cross section

The cross section for an SEU in the FPGA is given by σSEU = NSEU Φ · Nbits , where

◮ NSEU = Number of SEUs during irradiation, ◮ Φ = Particle fluence (number of incident particles per cm2), ◮ Nbits = Number of configuration-memory bits.

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 8/21 Markus Preston - Stockholm University

Experimental results - Neutron irradiation

Good agreement with previous measurements on this device:

The Svedberg Laboratory (TSL) Los Alamos Neutron Science Center (LANSCE) 4 6 8 10 12 14 SEU Cross Section [10−15 cm2 bit−1] This work [1] [2] [3] [4]

Cross section assuming only neutrons > 10 MeV cause SEU

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 9/21 Markus Preston - Stockholm University

Experimental results - Proton irradiation

Good agreement with previous measurements on this device:

60 80 100 120 140 160 180 Proton energy [MeV] 2 4 6 8 10 12 14 SEU Cross Section [10−15 cm2 bit−1] This work [5] [6] [7]

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 10/21 Markus Preston - Stockholm University

Experimental results - Proton irradiation

Good agreement with previous measurements on this device:

60 80 100 120 140 160 180 Proton energy [MeV] 2 4 6 8 10 12 14 SEU Cross Section [10−15 cm2 bit−1] This work [5] [6] [7]

Procedure validated!

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 10/21 Markus Preston - Stockholm University

What does this mean for PANDA?

From neutron and proton irradiations σSEU determined at specific energies. Conditions during PANDA operation Continuous energy spectrum of particles at front-end board location. SEU rate = ?

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 11/21 Markus Preston - Stockholm University

What does this mean for PANDA?

From neutron and proton irradiations σSEU determined at specific energies. Conditions during PANDA operation Continuous energy spectrum of particles at front-end board location. SEU rate = ? Needed: Way to predict SEU rate in PANDA, using results from irradiations.

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 11/21 Markus Preston - Stockholm University

What does this mean for PANDA?

From neutron and proton irradiations σSEU determined at specific energies. Conditions during PANDA operation Continuous energy spectrum of particles at front-end board location. SEU rate = ? Solution: Monte Carlo simulations 1) GEANT4 simulation of energy deposits in microelectronics ⇒ σSEU(E) 2) pandaROOT MC simulation of PANDA setup ⇒ Φ(E)

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 11/21 Markus Preston - Stockholm University

GEANT4 simulation of energy deposition in silicon

FPGA=Silicon-based microelectronics device Particle beam into Si ⇒ Energy deposits One memory cell contains Four cubic Sensitive Volumes (size: d3)

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 12/21 Markus Preston - Stockholm University

Energy deposition by protons

How much energy is deposited by 100 MeV protons in one of these Sensitive Volumes?

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 13/21 Markus Preston - Stockholm University

Energy deposition by protons

How much energy is deposited by 100 MeV protons in one of these Sensitive Volumes?

1 10 100 Energy deposit [keV] 100 101 102 103 104 105 Counts per bin

Electronic stopping of p in Si Nuclear reactions of p in Si

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 13/21 Markus Preston - Stockholm University

Energy deposition by protons

How much energy is deposited by 100 MeV protons in one of these Sensitive Volumes?

1 10 100 Energy deposit [keV] 100 101 102 103 104 105 Counts per bin

Critical Energy: Ecrit Edep > Ecrit ⇒ SEU

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 13/21 Markus Preston - Stockholm University

Energy deposition by protons

How much energy is deposited by 100 MeV protons in one of these Sensitive Volumes?

1 10 100 Energy deposit [keV] 100 101 102 103 104 105 Counts per bin

Critical Energy: Ecrit Edep > Ecrit ⇒ SEU σSEU, sim ∝ NSEU

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 13/21 Markus Preston - Stockholm University

Optimising the parameters

For a particular incident particle type and energy, the cross section σSEU, sim depends on:

◮ The Sensitive Volume size d ◮ The Critical Energy Ecrit

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 14/21 Markus Preston - Stockholm University

Optimising the parameters

For a particular incident particle type and energy, the cross section σSEU, sim depends on:

◮ The Sensitive Volume size d ◮ The Critical Energy Ecrit

Task: find (d, Ecrit) giving best agreement with all measured data

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 14/21 Markus Preston - Stockholm University

Optimising the parameters

For a particular incident particle type and energy, the cross section σSEU, sim depends on:

◮ The Sensitive Volume size d ◮ The Critical Energy Ecrit

Task: find (d, Ecrit) giving best agreement with all measured data Method:

◮ Simulate particle beams matching experiments. This gives

◮ Neutrons: σSEU, sim(d, Ecrit; TSL neutron spectrum) ◮ Protons: σSEU, sim(d, Ecrit; 80, 100, 184 MeV)

◮ Find (d, Ecrit) through a χ2 global fit of σSEU, sim to σSEU

from the experiments.

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 14/21 Markus Preston - Stockholm University

Simulation results

Result: d = 125 ± 6 nm, Ecrit = 4.6 ± 0.2 keV. Using these parameter values, good agreement with both proton and neutron data.

70 100 130 160 190 Proton energy [MeV] 2 4 6 8 10 12 14 SEU Cross Section [10−15 cm2 bit−1]

a)

The Svedberg Laboratory (TSL) neutron spectrum (>10 MeV)

b)

Experiment Simulation

PRELIMINARY

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 15/21 Markus Preston - Stockholm University

Energy-dependence in neutron cross section

Using the optimal (d, Ecrit) values, simulate neutrons at different energies ⇒ Energy dependence of σSEU for neutrons.

20 40 60 80 100 120 140 Neutron energy [MeV] 0.5 1.0 1.5 2.0 2.5 3.0 3.5 SEU Cross Section [10−15 cm2 bit−1]

PRELIMINARY

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 16/21 Markus Preston - Stockholm University

Energy-dependence in neutron cross section

Using the optimal (d, Ecrit) values, simulate neutrons at different energies ⇒ Energy dependence of σSEU for neutrons.

20 40 60 80 100 120 140 Neutron energy [MeV] 0.5 1.0 1.5 2.0 2.5 3.0 3.5 SEU Cross Section [10−15 cm2 bit−1]

PRELIMINARY

What is the neutron spectrum at the PANDA front-end?

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 16/21 Markus Preston - Stockholm University

pandaROOT simulation

Analysis and Monte Carlo simulation framework for PANDA. Includes detector geometry, p¯ p interaction, interaction products... Dual Parton Model (DPM) event generator used.

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 17/21 Markus Preston - Stockholm University

pandaROOT simulation

Analysis and Monte Carlo simulation framework for PANDA. Includes detector geometry, p¯ p interaction, interaction products... Dual Parton Model (DPM) event generator used. Assuming PANDA start-up luminosity and p¯

p = 1.5 GeV/c ⇒

200 400 600 800 1000 1200 1400 Neutron energy [MeV] 10−4 10−3 10−2 10−1 100 101 102 Neutron flux per 5 MeV [cm−2 s−1]

PRELIMINARY

Total neutron flux at EMC Forward Endcap front-end board: 100 cm−2 s−1

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 17/21 Markus Preston - Stockholm University

pandaROOT simulation

Analysis and Monte Carlo simulation framework for PANDA. Includes detector geometry, p¯ p interaction, interaction products... Dual Parton Model (DPM) event generator used. Assuming PANDA start-up luminosity and p¯

p = 1.5 GeV/c ⇒

200 400 600 800 1000 1200 1400 Neutron energy [MeV] 10−4 10−3 10−2 10−1 100 101 102 Neutron flux per 5 MeV [cm−2 s−1]

PRELIMINARY

Total neutron flux at EMC Forward Endcap front-end board: 100 cm−2 s−1 Currently, only done for neutrons

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 17/21 Markus Preston - Stockholm University

What this means for PANDA

GEANT4 simulation Energy dependence of σSEU for neutrons. pandaROOT simulation PANDA neutron spectrum at front-end boards.

+

Mean Time Between Failures in FPGA configuration due to neutrons in PANDA Forward Endcap EMC: 47 ± 10 hours per FPGA (∼ 400 FPGAs in total in Forward Endcap) Assumptions: Antiproton momentum = 1.5 GeV/c Luminosity = 1031 cm−2 s−1

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 18/21 Markus Preston - Stockholm University

Summary

◮ Experimental results agree with previous measurements, both

for neutrons and protons.

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 19/21 Markus Preston - Stockholm University

Summary

◮ Experimental results agree with previous measurements, both

for neutrons and protons.

◮ GEANT4 simulation of energy deposits in nanometric silicon

volumes has been developed:

◮ Four cubic Sensitive Volumes per memory cell (SV side = d) ◮ Simulate neutron and proton beams ◮ Energy deposit > Ecrit ⇒ SEU Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 19/21 Markus Preston - Stockholm University

Summary

◮ Experimental results agree with previous measurements, both

for neutrons and protons.

◮ GEANT4 simulation of energy deposits in nanometric silicon

volumes has been developed:

◮ Four cubic Sensitive Volumes per memory cell (SV side = d) ◮ Simulate neutron and proton beams ◮ Energy deposit > Ecrit ⇒ SEU ◮ Optimal parameter values found to give agreement between

experimental and simulated cross section for neutrons and protons:

◮ d = 125 ± 6 nm ◮ Ecrit = 4.6 ± 0.2 keV Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 19/21 Markus Preston - Stockholm University

Summary

◮ Experimental results agree with previous measurements, both

for neutrons and protons.

◮ GEANT4 simulation of energy deposits in nanometric silicon

volumes has been developed:

◮ Four cubic Sensitive Volumes per memory cell (SV side = d) ◮ Simulate neutron and proton beams ◮ Energy deposit > Ecrit ⇒ SEU ◮ Optimal parameter values found to give agreement between

experimental and simulated cross section for neutrons and protons:

◮ d = 125 ± 6 nm ◮ Ecrit = 4.6 ± 0.2 keV

◮ pandaROOT Monte Carlo simulation ⇒ Neutron flux in

PANDA at start-up conditions.

◮ Mean Time Between Failures in Configuration Memory due to

neutrons: 47 ± 10 hours per FPGA (∼ 400 FPGAs in Forward Endcap).

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 19/21 Markus Preston - Stockholm University

Outlook

◮ Further irradiations are planned: thermal neutrons, 14 MeV

neutrons

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 20/21 Markus Preston - Stockholm University

Outlook

◮ Further irradiations are planned: thermal neutrons, 14 MeV

neutrons

◮ Develop pandaROOT simulations

◮ Estimate flux of protons in EMC ◮ Estimate fluxes in barrel part of EMC ◮ Investigate cases with higher beam momentum and higher

luminosity

◮ Investigate error rates in other parts of the FPGA ◮ Analyse probabilities for multi-bit vs. single-bit upset, and

consequences for failure rates in PANDA

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 20/21 Markus Preston - Stockholm University

Outlook

◮ Further irradiations are planned: thermal neutrons, 14 MeV

neutrons

◮ Develop pandaROOT simulations

◮ Estimate flux of protons in EMC ◮ Estimate fluxes in barrel part of EMC ◮ Investigate cases with higher beam momentum and higher

luminosity

◮ Investigate error rates in other parts of the FPGA ◮ Analyse probabilities for multi-bit vs. single-bit upset, and

consequences for failure rates in PANDA

◮ Based on conclusions of these analyses: investigate error

mitigation methods for use in PANDA.

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 20/21 Markus Preston - Stockholm University

Thank you for your attention!

Special thanks go to the technical staff and accelerator operators at TSL, Uppsala and KVI-CART, Groningen

Measurements and Simulations of Single-Event Upsets in a 28-nm FPGA 21/21 Markus Preston - Stockholm University