Tools / Single events / DRAM stuck bits / Mechanism in 3D

What does one particle damage to make a DRAM cell stick?

Not the capacitor and not the well as a whole. A single proton or heavy ion leaves a recoil cascade in the silicon; the vacancy complexes that survive it become generation centers, and when they sit in the depletion region of the storage node junction, next to the gate edge where the field is highest, trap-assisted tunneling multiplies their emission until the node is drained before the next refresh. The view below walks through the strike, the cascade, the defect chemistry, the leakage and the anneal in a generic buried-word-line cell, with a small model that puts numbers on each step.

3Dgeneric 1T1C cell, buried word lineproton or heavy iontoy model with stated assumptionsruns in your browserStuck-bit mission estimatorEstimator helpNAND flash bit flips in 3D

Drag to turn, scroll or pinch to zoom. Step through the six stages with the buttons at the bottom left; the panel at the right sets the particle, where the cascade lands, how many centers survive, the temperature and the refresh period, and reads out the leakage against the refresh.About 0.7 MB, no external libraries.

The six stages of the view

1

Steady state

A written 1 sits near Vcore on the storage node. Thermal generation in the junction depletion region leaks it slowly toward the p-well; the refresh rewrites it long before the sense threshold is reached.

2

Strike

The ionization column of the proton or ion is collected or recombines within nanoseconds. It may upset the stored value once; it leaves no lasting change in the silicon.

3

Recoil cascade

A nuclear elastic or inelastic collision (proton) or a rare large-angle Coulomb scatter (heavy ion) knocks a silicon atom from its site. The recoil deposits its energy within a few nanometers, leaving vacancies and interstitials.

4

Defect evolution

Most pairs recombine within microseconds. Surviving vacancies migrate at room temperature and are captured by oxygen or by each other, forming VO, V2 and related complexes with levels in the band gap.

5

Leakage

A center in the depletion region emits electron-hole pairs that the field separates: electrons into the node, holes into the well. At the gate edge the field adds trap-assisted tunneling, and the node is drained below the threshold before the refresh. The cell reads as discharged on every cycle.

6

Anneal

The complexes dissociate with a spread of activation energies. A hold at temperature removes those below the demarcation energy E* = kT ln(νt); the leakage falls and the cell recovers. This is the recovery the mission estimator carries.

Where the damage is, and where it is not

The element that is damaged is the silicon of the reverse-biased junction between the n-type storage node diffusion, the diffusion on which the capacitor contact lands, and the p-type array well. More precisely it is the depletion region of that junction, and within it the high-field portion at the edge of the access transistor gate, the same region that limits retention through gate-induced drain leakage in an unirradiated cell. One or a few displacement defects placed there become field-enhanced generation centers whose current exceeds the retention budget of the cell. The storage capacitor is what gets discharged; it is not what gets damaged.

The early single-event hard error literature attributed stuck bits to micro-dose, the ionizing dose a single heavy ion deposits in the gate oxide of the access transistor [13]. Edmonds and Scheick reversed that interpretation for the Hyundai 16M × 4 SDRAM, arguing that the mechanism is displacement damage from a single particle creating a leakage current that drains the storage capacitor [4], and later work on newer parts has converged on the same picture: the generation of defects in the depletion region between the access transistor and the storage capacitor is what makes a stuck bit [5, 10]. Electron irradiation is the cleanest discriminator, since a single electron deposits negligible micro-dose; electron-induced stuck bits were nevertheless observed and attributed to large displacement damage clusters [9]. The DRAM reliability community located the same spot independently through the variable retention time problem: Umeda and colleagues used electrically detected magnetic resonance to argue that a single silicon vacancy-oxygen complex embedded near the surface at the drain-gate boundary of the cell is the metastable defect behind it [7], and Goiffon and colleagues showed that irradiation reproduces the phenomenon, with gamma rays producing interface centers and neutrons producing bulk centers in the junction depletion region [8].

The capacitor dielectric is not the site. An amorphous or nanocrystalline high-k stack does not form the discrete band-gap generation centers that crystalline silicon does under displacement, and leakage through it would run to the common plate at Vcore/2, pulling both data states toward the sense midpoint rather than toward one state. Observed stuck bits read back the same value on every cycle, the one corresponding to the discharged node [10], which is the signature of leakage to the p-well. The well as a doped volume is not damaged in any functional sense either: a defect in the neutral p-well, outside the depletion region, generates and recombines in balance and adds no current to the node. Only the defects that land inside the few tens of nanometers of depleted silicon at the well boundary of the storage node matter, and generally only those near the gate edge matter at room temperature.

Why the location matters: the arithmetic

The view's model is deliberately simple so that each step can be checked. Assume a cell capacitance of 20 fF, a core voltage of 1.1 V, and a sense amplifier that fails once the node has lost half its charge, so the charge margin is 11 fC. With a 64 ms refresh the critical leakage is 11 fC / 64 ms, about 0.17 pA, or about 106 electrons per second.

QuantityValueBasis
Charge margin Qm11 fCCs Vcore / 2 with 20 fF and 1.1 V
Critical leakage at 64 ms0.17 pA, about 1.1 × 106 e/sQm / tREF
One midgap center, zero field, 25 °Cabout 40 /sσ vth ni / 2 with σ = 10-15 cm2, vth = 107 cm/s, ni = 1010 cm-3
Same center at 85 °Cabout 2300 /sni(T) rises about 50-fold from 25 °C to 85 °C
Shortfall, thermal emission only500 to 20,000 timescritical rate over one center; a single cascade leaves at most tens of stable defects
Field enhancement at 0.95 MV/cm, 25 °Cabout 104Hurkx trap-assisted tunneling, m* = 0.25 m0, FΓ about 0.37 MV/cm at 300 K [11]
Field enhancement at 0.3 MV/cm, 25 °Cabout 10same expression; the enhancement is exponential in (F/FΓ)2

Thermal generation from the handful of defects a single recoil leaves is short of the critical rate by two to four orders of magnitude. The deficit is closed only by electric-field enhancement of the emission, and the DRAM VRT literature has established that trap-assisted tunneling plays an important role in cell leakage [8]. A field of that size exists only at the gate-edge corner of the storage node junction, where the off-state word line, often held negative, faces a node near Vcore across the gate oxide. In the view, three centers at the gate edge stick the cell at 25 °C, one center suffices at 85 °C, and the same centers moved to the depletion edge or into the neutral well do nothing. Real cells differ in capacitance, field profile, trap depth and defect chemistry, and there are generally exceptions in radiation effects; the point of the arithmetic is the sensitivity to position, not the particular numbers.

What this means in test and in flight

Several features that are seen in test follow from metastable defect complexes in a high-field region. Intermittency: Chugg and colleagues showed that intermittently stuck bits behave as a random telegraph signal, with the sticking and unsticking attributed to thermally driven fluctuations of the leakage through displacement damage complexes in depletion regions [6], and Goiffon and colleagues argued that VRT, a structural fluctuation of the complex without a change of charge state, is the most probable cause of intermittent stuck bits in irradiated DRAMs [8]. Annealing: Söderström and colleagues found that the largest radiation-induced retention losses anneal at room temperature while the least-affected bits continued to lose retention during the anneal [10], which is consistent with vacancy migration and complex formation after the hit rather than with the neutralization of oxide charge. Temperature and refresh: the per-center generation rate rises steeply with temperature while the field enhancement falls somewhat, and the stuck-bit count read from a part depends on the refresh period used to read it; the estimator's help page describes how its reads are tied to the flight temperature through the refresh period.

The stuck-bit mission estimator on this site takes the counts from such a test, their creation cross section per bit and their persistence in demarcation energy, and returns bit errors against mission duration. The anneal stage of the view is the same demarcation-energy picture applied to one cell. Generally this holds for stacked-capacitor cells with a buried word line; trench-capacitor architectures move the critical junction to the buried-strap outdiffusion, but the principle is the same.

The model in the view

  • Geometry: a generic 1T1C cell with a buried word line, a storage node contact and a stacked capacitor, drawn to explain and not to the scale of any product. The neighbor cell shares the bit line contact.
  • Cell: Cs = 20 fF, Vcore = 1.1 V, failure at loss of half the stored charge, retention time tret = Qm/I, stuck when tret < tREF. The node is discharged at constant current, a simplification of the bias-dependent leakage of a real junction.
  • Intrinsic leakage set to give 2 s retention at 85 °C, halving per 10 °C cooler.
  • Per-center generation g0 = σ vth ni/2 for a midgap center in a depleted region, with σ = 10-15 cm2, vth = 107 cm/s at 300 K and ni(T) from Eg = 1.12 eV; field enhancement 1 + Γ with Γ = 2√(3π)(F/FΓ) exp((F/FΓ)2) after Hurkx [11], m* = 0.25 m0; a guessed field map of 0.95 MV/cm along the gate edge falling linearly to zero at the depletion edge; nothing from centers in the neutral well.
  • Metastable defect: two configurations, the second with 1/50 of the emission rate, exponentially distributed dwell times compressed for the animation.
  • Anneal: each center has an activation energy drawn uniformly from 1.0 to 1.6 eV; a hold of t at T removes those below E* = kT ln(νt) with ν = 1013 /s.
  • Shown, not computed: the number and paths of the carriers, the size and shape of the cascade, and the timing of the stages.

This page is an explainer. Its numbers come from the model listed above and are not a qualification result for any part or mission; see the site disclaimer.

References

  1. G. M. Swift, D. J. Padgett and A. H. Johnston, "A new class of single event hard errors," IEEE Transactions on Nuclear Science, vol. 41, no. 6, pp. 2043 to 2048, 1994.
  2. L. D. Edmonds, S. M. Guertin, L. Z. Scheick, D. Nguyen and G. M. Swift, "Ion-induced stuck bits in 1T/1C SDRAM cells," IEEE Transactions on Nuclear Science, vol. 48, no. 6, pp. 1925 to 1930, 2001.
  3. H. Shindou, S. Kuboyama, N. Ikeda, T. Hirao and S. Matsuda, "Bulk damage caused by single protons in SDRAMs," IEEE Transactions on Nuclear Science, vol. 50, no. 6, pp. 1839 to 1845, 2003.
  4. L. D. Edmonds and L. Z. Scheick, "Physical mechanisms of ion-induced stuck bits in the Hyundai 16M × 4 SDRAM," IEEE Transactions on Nuclear Science, vol. 55, no. 6, pp. 3265 to 3271, 2008.
  5. L. Scheick, S. Guertin and D. Nguyen, "Investigation of the mechanism of stuck bits in high capacity SDRAMs," IEEE Radiation Effects Data Workshop, pp. 47 to 52, 2008.
  6. A. M. Chugg, A. J. Burnell, P. H. Duncan, S. Parker and J. J. Ward, "The random telegraph signal behavior of intermittently stuck bits in SDRAMs," IEEE Transactions on Nuclear Science, vol. 56, no. 6, pp. 3057 to 3064, 2009.
  7. T. Umeda, K. Okonogi, K. Ohyu, S. Tsukada, K. Hamada, S. Fujieda and Y. Mochizuki, "Single silicon vacancy-oxygen complex defect and variable retention time phenomenon in dynamic random access memories," Applied Physics Letters, vol. 88, 253504, 2006.
  8. V. Goiffon, T. Bilba, T. Deladerrière, G. Beaugendre, A. Le Roch, A. Dion, C. Virmontois, J.-M. Belloir, M. Gaillardin, A. Jay and P. Paillet, "Radiation induced variable retention time in dynamic random access memories," IEEE Transactions on Nuclear Science, vol. 67, no. 1, 2020.
  9. D. Söderström, L. Matana Luza, H. Kettunen, A. Javanainen, W. Farabolini, A. Gilardi, A. Coronetti, C. Poivey and L. Dilillo, "Electron-induced upsets and stuck bits in SDRAMs in the Jovian environment," IEEE Transactions on Nuclear Science, vol. 68, no. 5, pp. 716 to 723, 2021.
  10. D. Söderström, L. Matana Luza, A. Martins Pio de Mattos, T. Gil, H. Kettunen, K. Niskanen, A. Javanainen and L. Dilillo, "Technology dependence of stuck bits and single-event upsets in 110, 72, and 63-nm SDRAMs," IEEE Transactions on Nuclear Science, vol. 70, no. 8, 2023.
  11. G. A. M. Hurkx, D. B. M. Klaassen and M. P. G. Knuvers, "A new recombination model for device simulation including tunneling," IEEE Transactions on Electron Devices, vol. 39, no. 2, pp. 331 to 338, 1992.
  12. J. R. Srour and J. W. Palko, "Displacement damage effects in irradiated semiconductor devices," IEEE Transactions on Nuclear Science, vol. 60, no. 3, pp. 1740 to 1766, 2013.
  13. T. R. Oldham, K. W. Bennett, J. Beaucour, T. Carriere, C. Polimadei and S. Sommer, "Total dose failures in advanced electronics from single ions," IEEE Transactions on Nuclear Science, vol. 40, no. 6, 1993.
  14. P. J. Restle, J. W. Park and B. F. Lloyd, "DRAM variable retention time," IEDM Technical Digest, 1992.

To the stuck-bit mission estimator · Estimator help and sources · NAND flash bit flips in 3D