Mainstream NAND flash memories are currently manufactured on 4xnm processes with major NAND flash vendors migrating to 3xnm this year. In the race to reduce costs, NAND flash manufacturers are developing 2xnm technology, however with performance and reliability characteristics severely degraded relative to the 4xnm generation, 2xnm floating gate NAND flash could be the last process technology generation. What’s next?
NAND flash vendors have been exploring a variety of alternatives including spin-torque MRAM, nanocrystal memory, phase change memory and resistive memory. However as lithographic scaling becomes more challenging, companies are turning their sights to vertically stacked implementations of memory cells or 3D memory. Among the candidates: stacked NAND technologies employing charge trapping technology, vertical memory cells etched in a pillar and stackable cross-point memory arrays.
It is the aim of this report to explore in detail the feasibility of each of these alternatives as a candidate to replace NAND flash memories within the next four years.
List of Figures
List of Tables
NAND Flash Memory
NAND Flash Memory Technology
3D Memory Alternatives
Samsung Stacked TANOS
Advantages and Disadvantages
Toshiba Bit-Cost Scalable NAND
Advantages and Disadvantages
Cross-point Memory Arrays
Comparison of 3D Memory Concepts
Key Parameters of stacked NAND
Key Parameters of vertical NAND (BiCS)
Key Parameters of Stacked Cross-Point Array Memory
About the Authors
About Forward Insights
List of Figures:
Figure 1. NAND Flash Technology Evolution
Figure 2. NAND Flash Memory Gap Fill
Figure 3. NAND Flash Memory Coupling Ratio and Cross-talk
Figure 4. Electrons Stored on the Floating Gate
Figure 5. Samsung 32Gb CTF Memory
Figure 6. Estimation of NAND fabrication costs as the bit density increases for conventional planar and stacked cell
Figure 7. Schematic cross section of the 3-D TANOS NAND showing a first arrangement of cell strings on Si substrate level and a second level of NAND memory cell strings stacked above
Figure 8. SEMs of 3D stacked NAND cell string. The 2nd active layer is SOI-like perfect single crystal
Figure 9. Layout and vertical structure of word-lines and x-decoders for the doubly stacked NAND flash memory cell array.
Figure 10. Key process steps of 3D NAND memory.
Figure 11. Comparison of erase operation (a) well bias initiated erase by block, (b) erase by page without well bias in the floating body case
Figure 12. (a) TEM cross section image of LEG Si film; Inset image shows sub-grain boundary in protrusion. (b) Modelling of LEG process; selective melting of a-Si film and solidification from seed.
Figure 13. Process scheme with amorphous layer formation on seeds to initiate laser epitaxial growth of single crystal Si film resulting in stacked transistor bodies for 3D memories.
Figure 14. Tilted SEM image with protrusions which are precisely located in the center between neighbouring seeds.
Figure 15. Tri-gate structure of the TFT device based SONOS memory cell (Macronix)
Figure 16. TEM cross sections of TFT NAND strings perpendicular to the wordline and along the wordline direction
Figure 17. Schematic illustration of thermal budget impact on device structure: S/D junctions between NAND wordlines diffuse below gate contacts and cause punchthrough device failure.
Figure 18. Simulated temperature profiles for various laser pulse durations. The penetration depths can be controlled by the pulse durations.
Figure 19. The basic concept of the DG-TFT-SONOS Flash showing simultaneous shielding of stored charge during pass voltage application and intimate electrical interaction for good short channel control.
Figure 20. XTEM along NAND string channel of the dual gate SONOS devices
Figure 21. SIMS analysis of antimony-implanted LPCVD amorphous silicon that subsequently underwent the illustrated thermal steps. This mimics the behavior of the source and drain dopant during processing. Negligible diffusion occurs.
Figure 22. Sheet resistance of antimony implanted into LPCVD amorphous silicon and annealed.
Figure 23. Cycling endurance of the mid-cell of a 32 cell string of minimum feature sized devices. No other second-gated memory devices are used in the cycling.
Figure 24. Retention after cycling 105 cycles showing good performance after extrapolation to 10 years.
Figure 25. Equivalent circuit of the BiCS arrangement of vertical NAND strings.
Figure 26. Birds-eye view of BiCS flash memory. NCG is the number of control gates. NSG is the number of rows of pillars sharing an upper select gate.
Figure 27. (a) Cross section of BiCS flash memory string, (b) Cross section of vertical SONOS cell
Figure 28. Stair-like structure at the edge of control gate plates and upper select gate pitch.
Figure 29. Fabrication sequence of BiCS flash memory
Figure 30. Cross sectional SEM image and schematic illustration of the latest reported BiCS cell array.
Figure 31. Id-Vg characteristics after program and erase operation of each of the different cells of a NAND string (non-optimized structure with respect to disturb).
Figure 32. Schematic potential profile in unselected pillars during program operation. The potential of pillars boosted by Vpgm is seen to be higher than Vpass and a negative Vth shift of the cell to be programmed is achieved.
Figure 33. Vth shift after program disturb of programmed cells in unselected pillars. A negative Vth shift is suppressed if suitable Vpass is selected.
Figure 34. Improved concept of changing lower select gate plates to line and space pattern in order to boost unselected pillars by turning on/off USG and LSG synchronously.
Figure 35. Vth shift after read disturb on unselected pillars. Vth shift is suppressed significantly by changing the lower select gate plates into line and space patterns.
Figure 36. Schematic illustration of SONS and SONONS memory layer scheme. The tunnel film in the SONONS structure inhibits the release of trapped charges of the trapping layer. Retention and disturb behavior is improved.
Figure 37. Program and erase characteristics of SONONS based BiCS cells.
Figure 38. Data retention results of SONONS and SONS cells of BiCS flash.
Figure 39. This Besang picture shows an SEM (right) of the vertical memory cells of a 10 kb flash memory array with schematics (left) depicting the stacked memory and CMOS circuitry
Figure 40. Schematic of a PCM memory array with nanowire diodes as memory cell select devices
Figure 41. (a) Shows a generalized cross-point memory structure whose one bit cell of the array consists of a memory element and a switch element between conductive lines on top (wordline) and bottom (bitline). (b) Illustration of reading interference in an array consisting of 2 × 2 cells without switch elements. (c) Rectified reading operation in an array consisting of 2 × 2 cells with switch elements
Figure 42. Matrix 3D poly Si anti-fuse diode cell schematics and electrical characteristics of programmed and unprogrammed cell
Figure 43. Matrix 3D poly Si anti-fuse diode cross section
Figure 44. Matrix 3D poly Si anti-fuse memory cross section
Figure 45. Cross section of the 512Mb PRAM with diodes as select devices
Figure 46. Process sequence for the vertical diode and the self aligned bottom electrode contact (SABEC)
Figure 47. Simulated temperature profile while writing RESET state. The temperature of the pn-junction in vertical diode is not too high to degrade the device operation
Figure 48. Expected evolution of cell structures: Cross sections of the PRAM switching elements
Figure 49. Simulated write current for melting at the phase transformation core of (a) a conventional planar PCM and (b) a confined cell structure. When the reset current is applied to the confined cell structure, melting of PCM is expected to occur between electrodes limited to isolated cell.
Figure 50. TEM images of dash-type confined cell structure. Even with 7.5nm width: The PCM was filled perfectly without voids
Figure 51. Endurance characteristic of confined PCM cell.
Figure 52. Proposed resistance switching mechanism of RRAM. The Schottky barrier is reversibly modulated by charging/discharging of vacancies
Figure 53. Left: Schematic potential energy diagram of the TaOx redox pair: ? G is the smallest of all commonly known resistive switching material candidates indicating that both HRS and LRS are stable. Right: Schematic change of the barrier height corresponding to the redox reaction according to the model assumption
Figure 54. Cross section of fabricated 1T1R ReRAM cells based on 0.18µm CMOS process.
Figure 55. (a) Workfunction versus electrode potential of various electrode metals and appearance of the switching effect, (b) switching behaviour with Ir electrode and (c) with Pt electrode
Figure 56. (a) I-V plot of the 1T1R memory cell (cell area 0.5 x0.5 µm²) in pulse voltage sweep with 100ns pulse width, (b) Schottky plot of I-V characteristic
Figure 57. Endurance properties of ReRAM (left) and data retention of TaOx memory cells at 150°C (right)
Figure 58. Sketch of the cell cross section (top) and sample cross section of the fabricated die with enlarged TEM (right),
Figure 59. I-V characteristic of a memory cell in sweeping mode
Figure 60. Read disturb characteristics for set and reset status with -0.1 V bias forcing to reset and +0.1 V bias forcing to set, respectively. No degradation is found up to 103 seconds
Figure 61. Scaling characteristics of resistance distributions for (a) 50 nm diameter cells and (b) 20 nm diameter cells without verification loop. The inset shows an AFM image of a 20 nm diameter cell
Figure 62. Schematic diagram of a 2-stack 1D-1R cell arrangement with upper layers reversed to share the bitline. The line width of word and bitlines was 0.5µm resulting in a final cell size of 0.5x0.5µm²
Figure 63. Conceptual schematic diagram of a stacked memory with TFTs decoders as stackable peripheral circuits
Figure 64. Specifications for 43nm SLC and MLC NAND Flash
Figure 65. Lithography Resolution Limits
Figure 66. Relative Cost per Bit of Stacked SLC NAND
Figure 67. Relative Cost per Bit of Stackable Non-volatile Memories
List of Tables:
Table 1. Multi-bit NAND Flash Comparison
Table 2. Comparison of Multilayer NAND with SLC NAND
Table 3. Comparison of Multilayer NAND with MLC NAND
Table 4. Comparison of BiCS NAND with SLC and MLC NAND
Table 5. Comparison of Stacked Cross Point Array Memory with SLC and MLC NAND
Josef brings with him a wealth of research and development experience in semiconductor memories from 26 years at Siemens Semiconductor / Infineon Technologies / Qimonda AG. Prior to joining Forward Insights, Josef was a principal at Qimonda Flash GmbH responsible for evaluating patents and intellectual property and developing innovative non-volatile memory technology and novel cell concepts to overcome the ultimate technology scaling constraints. Josef was instrumental in enabling the start-up company Ingentix to successfully demonstrate the feasibility of charge trapping technology for mass storage products.
In addition to his technical publications, Josef holds numerous memory-related patents and was named Infineon’s Inventor of the Year in 2004. He has been a member of the technical committee of the ICMTD (International Conference on Memory Technology and Design) for the past three years.
Josef holds a Dr. rer. nat. from the Technical University in Munich in solid state physics.
Gregory Wong is the Founder and Principal Analyst. Greg has in-depth knowledge of the cost, performance and markets and applications of 2-bit per cell NOR, NROMand NAND flash, 3-bit per cell and 4-bit per cell NAND and 4-bit per cell NROM flash technologies as well as solid state drives. Greg has authored a variety of reports pertaining to the technology, performance, costs, markets and applications of flash and alternative non-volatile memories, solid state drives, 3-bit and 4-bit per cell NAND flash memories, flash cards and removable storage, and embedded flash memories. He also tracks wafer capacity and shipments by vendor and issues quarterly supply-demand and capex forecasts.
Greg has 11 years of management experience in strategic planning, business development and engineering at Hitachi, Siemens, ProMOS and Qimonda/Infineon. In these positions, Greg was responsible for analyzing and evaluating flash memory vendors’ strategies, process technologies, design architectures, product performance, manufacturing capabilities and costs. Greg worked closely with senior management on strategy formulation and setting operational performance targets. He has worked in Canada, Japan, China, Taiwan and Germany.
Greg earned his B.A.Sc. degree in Electrical Engineering from the University of Toronto, and his M.B.A. degree from the Richard Ivey School of Business in London, Ontario.