Reliability issues and modeling of Flash and post-Flash memory (Invited Paper)

Flash memory, in particular NAND, has been an enabling technology for portable applications for the last two decades. The strength of Flash is its excellent scaling capability, allowing an ever increasing density at a decreasing cost and maintained reliability. However, the geometrical scaling of the cell exacerbates charge loss and fluctuation effects. On the other hand, new post-Flash memory technologies are being proposed, with different storage concepts and reliability physics. This review discusses the major reliability issues for Flash, with emphasis on the physical mechanisms and modeling. The reliability of charge trap and resistive memories, such as phase change and resistive switching memories, is addressed.     

Defect generation statistics in thin gate oxides

We analyze data-retention experiments for flash memory arrays with thin tunnel oxide (t/sub ox/ = 5 nm). These samples show an additional conduction mechanism besides Fowler-Nordheim tunneling and stress-induced leakage current (SILC). The additional leakage contribution is analyzed with respect to the spatial distribution in the array and the shape of the current-voltage characteristics, and is interpreted as an anomalous SILC due to a two-trap leakage path. From the cycling dependence of the distribution tails related to one- and two-trap leakage, we provide evidence that the defect generation statistics is not Poissonian, but is instead correlated. Possible physical mechanisms responsible for correlated generation are also discussed.     

A statistical model for SILC in flash memories

The reliability of flash memories is strongly. limited by the stress-induced leakage current (SILC), which leads to accelerated charge-loss phenomena in a few anomalous cells. Estimating the reliability of large flash arrays requires that physically-based models for the statistical distribution of SILC are developed. In this paper, we show a physical model for the leakage mechanism in thin oxides, which is able us to explain the anomalous leakage-conduction in tail cells. The physical model is then used for a quantitative evaluation of the SILC distribution in large flash arrays. The new model can reproduce the statistics of SILC for a wide range of tunnel-oxide thickness, and can provide a straightforward estimation of the reliability for large flash arrays.     

Parasitic reset in the programming transient of PCMs

We studied the programming dynamics in phase change memory cells. It is shown that programming in stand-alone cells is strongly affected by the parasitic capacitance in the measurement setup, leading to a current overshoot after threshold switching of the amorphous chalcogenide. This results in a parasitic melting and quenching of the active material, affecting the current distribution during program and the final phase distribution in the active material. The relevance of this artefact for real-device operation is discussed with reference to the value of the parasitic capacitance.     

Cyclic beta-(1,2)-glucan synthesis in Rhizobiaceae: roles of the 319-kilodalton protein intermediate

Cyclic beta-(1,2)-glucans are synthesized by members of the Rhizobiaceae family through protein-linked oligosaccharides as intermediates. The protein moiety is a large inner membrane molecule of about 319 kDa. In Agrobacterium tumefaciens and in Rhizobium meliloti the protein is termed ChvB and NdvB, respectively. Inner membranes of R. meliloti 102F34 and A. tumefaciens A348 were first incubated with UDP-[14C]Glc and then solubilized with Triton X-100 and analyzed by polyacrylamide gel electrophoresis under native conditions. A radioactive band corresponding to the 319-kDa protein was detected in both bacteria. Triton-solubilized inner membranes of A. tumefaciens were submitted to native electrophoresis and then assayed for oligosaccharide-protein intermediate formation in situ by incubating the gel with UDP-[14C]Glc. A [14C]glucose-labeled protein with an electrophoretic mobility identical to that corresponding to the 319-kDa [14C]glucan protein intermediate was detected. In addition, protein-linked radioactivity was partially chased when the gel was incubated with unlabeled UDP-Glc. A heterogeneous family of cyclic beta-(1,2)-glucans was formed upon incubation of the gel portion containing the 319-kDa protein intermediate with UDP-[14C]Glc. A protein with an electrophoretic behavior similar to the 319-kDa protein intermediate was "in gel" labeled by using Triton-solubilized inner membranes of an A. tumefaciens exoC mutant, which contains a protein intermediate without nascent glucan. These results indicate that initiation (protein glucosylation), elongation, and cyclization were catalyzed in situ. Therefore, the three enzymatic activities detected in situ reside in a unique protein component (i.e., cyclic beta-(1,2)-glucan synthase). It is suggested that the protein component is the 319-kDa protein intermediate, which might catalyze the overall cyclic beta-(1,2)-glucan synthesis.     

Voltage-Controlled Relaxation Oscillations in Phase-Change Memory Devices

A new oscillation behavior in a phase-change memory device is presented and analyzed. The device consists in a chalcogenide resistor with a parallel capacitance and no inductance. Biasing the device immediately after a proper trigger pulse leads to damped relaxation oscillations, which can be controlled in frequency by the bias voltage. The oscillation mechanism is explained by repetitive cycles of threshold switching and recovery of the high-resistance (off) state of the amorphous chalcogenide region in the device. Damping is explained by oscillation-induced phase change in the chalcogenide layer.     

Modeling of Programming and Read Performance in Phase-Change Memories—Part I: Cell Optimization and Scaling

One of the major concerns for the feasibility of phase-change memories is the reduction of the programming current. To this aim, several efforts have been dedicated both on cell geometry and on material engineering. This paper addresses programming-current minimization by the optimization of the cell geometry and materials, programming-current scaling, and the tradeoff between programming and readout performances of the cell. A general procedure to find the optimum-cell geometry is proposed and applied to a prototype vertical cell. Then, the evolution of program and read performances through technology nodes is analyzed by numerical simulations with the aid of an analytical model, for both the isotropic- and nonisotropic-scaling approaches. The two scaling approaches are discussed and compared in terms of program and read cell performances. Finally, material optimization is considered for further program-read improvement.     

Modeling of SILC based on electron and hole tunneling. I. Transienteffects

A detailed investigation of the steady-state and transient leakage currents in thin oxides is proposed. The experimental data are compared with numerical results obtained from a model based on an inelastic trap-assisted tunneling process, which includes both electron and hole contributions. In order to accurately reproduce the transient discharge currents, a continuous distribution of oxide traps was adopted. The energies of these levels can be either in correspondence of the conduction or valence band edges of the adjacent silicon/polysilicon layers. Both electrons and holes contribute to the transient stress-induced leakage current (SILC), but the extracted trap densities cannot account for the steady-state SILC. A different mechanism, involving trap levels with energy aligned to the energy gap of the silicon layers is proposed and is developed in the following paper. The model can be applied to any type of device and bias conditions and may be used to correctly recognize the role of electron and hole SILC and the spatial and energy distribution of defect states     

Modeling of SILC based on electron and hole tunneling. II.Steady-state

For pt. I see ibid., vol. 47, no. 6 (June 2000). A numerical model for the stationary stress-induced leakage current (SILC) is presented, accounting for both electron and hole tunneling. Detailed comparisons against experimental results on both n- and p-channel devices highlight that the steady-state SILC is due to positively charged centers, with an energy level located in correspondence of the silicon bandgap. Electron-hole recombination at these sites dominates normal trap-assisted tunneling at low oxide fields, and successfully accounts for recently observed hole steady-state leakage. The contribution from neutral traps seems instead marginal. Based on this new picture, the impact of the recombination process on the leakage properties of ultrathin gate is also discussed