Gouri Sankar Kar

10×10nm 2 Hf/HfO x crossbar resistive RAM with excellent performance, reliability and low-energy operation

We report on worlds smallest HfO2-based Resistive RAM (RRAM) cell to date, featuring a novel Hf/HfOx resistive element stack, with an area of less than 10×10 nm2, fast ns-range on/off switching times at low-voltages and with a switching energy per bit of <0.1pJ. With excellent endurance of more than 5.107cycles, large on/off verified-window (>50), no closure of the on/off window after 30hrs/200C and failure-free device operation after 30hrs/250C thermal stress, the major device-level nonvolatile memory requirements are met. Furthermore, we clarify the impact of film crystallinity on cell operation from a scalability viewpoint, the role of the cap layer and bring insight into the switching mechanisms.

Endurance/Retention Trade-off on

The endurance/retention performance of HfO2/ Metal cap RRAM devices in a 1T1R configuration shows metal cap dependence. For Hf and Ti caps, owning strong thermodynamic ability of extracting oxygen from HfO2, long pulse endurance (>1010 cycles) could be achieved. For Ta cap, owning lower thermodynamic ability of extracting oxygen from HfO2, better retention can be achieved. Therefore, an endurance/retention performance tradeoff is identified on the 40 nm × 40 nm HfO2/Metal cap bipolar RRAM devices. The tradeoff of endurance/retention performance can be explained by a different filament constriction shape depending on metal cap layer as derived from fitting I-V curves in the quantum point contact model. This difference in filament constriction shape is attributed to the thermodynamics difference of metal cap: Hf and Ti have a stronger thermodynamical ability to extract oxygen from HfO2 than Ta. The possibility of tuning the intrinsic reliability performance by changing the cap materials paves a way for optimizing the operation of RRAM devices into the desired specifics.

\hbox{HfO}_{2}/\hbox{Metal}

By tuning the SET/RESET pulse amplitude conditions, the pulse endurance of our 40-nm HfO2/Hf 1T1R resistive-random-access-memory devices demonstrates varying failure behaviors after 106 cycles. For unbalanced SET/RESET pulse amplitude conditions, both low-resistance state (LRS) and high-resistance state (HRS) failures may occur, while varying the pulsewidths influences the LRS/HRS window and the stability of the LRS/HRS states. The failure of the HRS or LRS state during cycling is ascribed to the depletion or excess of oxygen vacancies at the switching interface. Through a dc SET/RESET recovery operation, LRS/HRS states can be recovered after failure, indicating that the distribution of oxygen vacancies can be restored. By optimally balancing the SET/RESET pulse conditions, more than 1010 pulse endurance cycles is achieved.

Cap 1T1R Bipolar RRAM

An analytic dynamic hour glass model for HfO2 RRAM is demonstrated, describing the reset as a dynamic equilibrium process and the set as a constriction growth limited by ion mobility and current compliance. The dependence on time, voltage and forming conditions is in good constriction growth agreement with experiments. Since the model is fully analytical, it can be implemented in a circuit simulator.

Balancing SET/RESET Pulse for

One of the key concerns related to low operating current (<;50μA) of RRAM is the degraded data retention. Most of the retention studies so far focused on high switching current range. In this work, we investigate the retention degradation mechanism at low programming current range (10-40μA) and identify the key parameters that control retention in oxygen vacancy filamentary switching HfO<;sub>2<;/sub>/Hf 1T1R RRAM cells. Based on this understanding we demonstrated significant improvement in retention by adding an additional thermal budget into our process flow. The impact of the Forming process on retention property was also investigated and Forming/SET conditions were optimized to improve the retention without increasing the operation current.

>\hbox{10}^{10}

Bipolar switching transition metal-oxide (TMO) RRAM devices are intensively studied as possible non-volatile memory for 1x nm node. HfO2 based stacks with excellent operation, good endurance and retention have been proposed [1, 2, 3], with demonstrated scalability down to <;10nm [3]. However, characterization of the reliability failure modes and understanding of the degradation mechanism is urgently needed, especially in the low operation current range relevant for practical application of RRAM devices. Although retention and endurance models in different TMO have been proposed [4, 5, 6], an in-depth understanding of endurance is still lacking for scaled HfO2 RRAM in low current operation.

Endurance in

A novel procedure to decompose the I- V switching curves of complementary resistive switching (CRS) RRAM cells into the intrinsic switching characteristics of its individual constituting elements is proposed based on the set behavior of HfO2-based bipolar RRAM elements. Analysis of different types of complementary cells indicates that very similar intrinsic switching behaviors occur in strongly different types of bipolar switching RRAM, however with a strong material dependence of the characteristic switching voltage.

\hbox{HfO}_{2}\hbox{/Hf}

A vertical cylindrical SONOS cell with a novel bilayer polysilicon channel down to 22-nm diameter for 3-D NAND Flash memory is successfully developed. We introduce a thin amorphous silicon layer along with the oxide-nitride-oxide (ONO) gate stack inside the memory hole. This silicon layer protects the tunnel oxide during opening of the gate stack at the bottom of the memory hole, after which it serves as the first layer of the bilayer polysilicon channel. This approach enables the 3-D architecture to achieve minimum cell area (4F2, with F being the feature size) without the need for the so-called pipeline connections. The smallest functional cells have the memory hole diameter F = 45 nm, resulting in 22-nm channel diameter. In case 16 cells are stacked, F = 45 nm would correspond to an equivalent 11-nm planar cell technology node. Excellent program/erase and retention obtained with the all-deposited ONO stack are demonstrated.

1T1R Bipolar RRAM

We optimize a 90nm-wide CuTe-based 1T1R CBRAM cell for highly controlled and ultrafast programming by engineering Al2O3 electrolyte and Ti buffer layers of appropriate density and thickness resp. By means of electrical and ab initio modeling, we demonstrate that switching is mainly controlled by field-driven motion of Cu+ species. Sub-ns programming is allowed by strong ionic-hopping barrier reduction over short insulating gap. Complete picture of conductance and switching phenomenology is shown in the entire operation range.

Dynamic ‘hour glass’ model for SET and RESET in HfO 2 RRAM

In this work, we present a detailed electrical characterization of TiN\HfO2\Hf\TiN RRAM elements, and show for the first time the intrinsic switching characteristics in the low current operation regime (100uA till few uAs) of small scaled cells (20nm) under DC and fast ramps (up to 1MV/s) condition, using a newly proposed 2R test structure. The main characteristic parameters of the SET and RESET switching are defined and their speed dependence is characterized. Resistance decrease during SET is observed to occur at a constant voltage Vtrans, while symmetrically RESET process starts at -Vtrans. With reducing operation current, mean values of the symmetric SET and RESET transition voltages remain unchanged but their spread strongly increases. The actual I-V characteristics show discrete current jumps and non-linear quantum-mechanical conduction is evidenced and more pronounced at smaller currents. Increasing the ramp rate increases the SET and RESET transition voltage logarithmically with a concomitant reduction of the HRS resistance. These measurements allow for a better insight and understanding of the dynamic switching properties in low-current, ultra-scaled RRAM cells.