Download presentation
Presentation is loading. Please wait.
1
R.W. Mann and N. George ECE632 Dec. 2, 2008
Circuit Design Methodologies for Soft Error Resilience in Nano- Scaled CMOS Technologies R.W. Mann and N. George ECE632 Dec. 2, 2008
2
OUTLINE Section 1 (RWM): Brief background and underlying physics
Methods for modeling and characterizing SEU Trends for nano-scaled SRAM Cell topology effects (node cap, node-node cap) Section 2 (NG): Inherent resilience of logic to SEU How much critical charge Footprint of a particle strike Logic Interleaving and Overheads Describe the 2 primary sources of radiation induced SEU - alpha -- source impurities in packaging materials -- most common source of radioactive trace elements is Pb ---most notably 210Po 206 Pb which emits a 4He alpha ---flux 0.001cm2/hr - neutron -- source is galactic subatomic particles with energy > few Gev --- less than a few GeV will be swept away by the earths magnetic field --- particles which penetrate the earths magnetic field undergo cascades of collisions within the earths atomosphere until reaching the surface where the flux of particles that can cause upset is primarily neutrons which have a flux of ~20 n/cm2/hr - concept of SEU and MBU (key points: soft error, single and multi bit, alpha and neutron, pulse modeling, funneling and diffusion, scaling, SRAM and interleaving) 2/22/2019 UVA ECE 632 final project
3
Background physics underlying SEU in microelectronics circuits
Section 1.1: Background physics underlying SEU in microelectronics circuits 2/22/2019 UVA ECE 632 final project
4
Long ago and far away 2/22/2019 UVA ECE 632 final project
5
Sources of Radiation induced upset
Two primary sources of radiation induced SEU Neutron particle Neutron flux of ~10-20 n/cm2/hr at sea level result of high energy proton collisions which penetrate the earths magnetic field Ionization by “indirect ionization” – creating (heavy and light secondary particles) e/hole charge cloud Mitigation limited (Shielding requires ~5ft thick concrete ) alpha particle (doubly ionized He nucleus) primary source -impurities in solder and packaging materials most common source of radioactive trace elements is Pb most notably 210Po 206 Pb which emits a 4He alpha (5.3MeV) flux 0.001cm2/hr Ionizing radiation “direct ionization” – creating e/hole pairs along path Mitigation strategies include increased material purity, design rules and on chip barriers (40um of Cu for example) 2/22/2019 UVA ECE 632 final project
6
SER F = particle flux Adiff= area of sensitive diffusion
Qs = charge collection efficiency f(doping profiles,Vdd) Qcrit = minimum charge required to flip the state Commonly characterized as FIT/Mb (number of failures for a one Mb device in 109 hr) MTTF (mean time to fail) = 109hr/(FIT/Mb * #Mb) 1000 FIT/Mb typical for SRAM MTTF for 256Mbyte ~500hr at sea level MTTF for 256Mbyte ~1.5hr in flight 2/22/2019 UVA ECE 632 final project
7
Section 1.2: Modeling charge collection and Qcrit using a circuit simulation tool: Spectre 2/22/2019 UVA ECE 632 final project
8
Simulation of charge collection event on sensitive node for SRAM
WL WL BL BLB 1 Double exponential Current pulse I t 2/22/2019 UVA ECE 632 final project
9
Trends in nano-scaled SRAM
Section 1.3: Trends in nano-scaled SRAM 2/22/2019 UVA ECE 632 final project
10
Comparison of Qcrit values for SRAM with previously published values
2/22/2019 UVA ECE 632 final project
11
Qcrit values for SRAM across available PTM technologies
2/22/2019 UVA ECE 632 final project
12
Effects of SRAM cell topology
Section 1.4: Effects of SRAM cell topology 2/22/2019 UVA ECE 632 final project
13
Effect of W and Cap on Qcrit
2/22/2019 UVA ECE 632 final project
14
Summary Two primary sources (alpha & neutron) account for majority of SEU Charge collection mechanism characterized by double exponential current pulse: simulated using circuit simulation tool Ocean script with binary search to find Qcrit used to explore effects of: Technology (LP vs HP) Scaling (130nm – 16nm) and Vdd Cell topology effects Capacitance (node and node-to-node) characterized both forms of capacitance additions show linear dependence Node-node capacitance more significant Device width evaluated across PTM technologies 0.3um width adder ~ 10fF adder to node capacitance for 32nm cell Qcrit (alpha pulse) linearly declining at rate of 0.023fC/nm Technology options limited: Circuit strategies for mitigation (such as ecc and interleaving) are becoming more necessary with continued scaling 2/22/2019 UVA ECE 632 final project
15
Section 2: What’s different about logic?
16
Known to be more resilient to soft errors than SRAM Masking effects
Logic and SEUs Known to be more resilient to soft errors than SRAM Masking effects Logical masking Electrical masking Latching-window masking Size of logic transistors ? D CLK Q 2/22/2019 UVA ECE 632 final project
17
Vulnerable Latching Window
Sequential elements can latch a glitch during setup and hold time 2/22/2019 UVA ECE 632 final project
18
Width of Vulnerable Latching Window
2/22/2019 UVA ECE 632 final project
19
Critical Charge 2/22/2019 UVA ECE 632 final project
20
Footprint of a particle strike
Interleaving Logic 2/22/2019 UVA ECE 632 final project
21
Overheads of Interleaving
4-Bit Brent-Kung Adder protected with parity 2/22/2019 UVA ECE 632 final project
22
Questions? Thank you 2/22/2019 UVA ECE 632 final project
23
Beyond here are Back up slides
2/22/2019 UVA ECE 632 final project
24
FILE LOCATIONS FOR SIMULATION OF CRITICAL CHARGE
Each of the following directories contain (1) ocean script – input.tpl and (2) Netlist which defines the pulse and SRAM circuit and (3) readme.txt ~SRAM_TOOL/device/TESTS/SRAM_Qc_neutron ~SRAM_TOOL/device/TESTS/SRAM_Qc_alpha The specific technology and corresponding Vdd and critical dimensions are Specified in the following file location ~SRAM_TOOL/template/ 2/22/2019 UVA ECE 632 final project
25
Carry Checking & Parity Prediction
2/22/2019 UVA ECE 632 final project
Similar presentations
