In December 2011, at the IEDM conference Hynix presented their version of the next generation of NAND flash (smaller than 20nm). Key developments presented: "A middle-1x nm design rule multi-level NAND flash memory cell (M1X-NAND) has been successfully developed for the first time.
1) Quad spacer patterning technology (QSPT) of ArF immersion lithography is used for patterning mid-1x-nm rule wordline (WL). In order to achieve high performance and reliability, several integration technologies are adopted, such as
2) advanced WL air-gap process,
3) floating gate slimming process, and
4) optimized junction formation scheme. And also, by using
5) new N±1 WL Vpass scheme during programming, charge loss and program speed are greatly improved."
See more details below.
A middle-1X nm NAND flash memory cell (M1X-NAND) with highly manufacturable integration technologies
Joowon Hwang, Jihyun Seo, et al., Hynix Semiconductor Inc.
5/14/2012 3:21 PM EDT
Editor’s note: This work was first presented at the 2011 IEEE International Electron Devices Meeting (IEDM) and appears here courtesy of the IEEE.
For more information about IEDM 2012 (San Francisco, CA; December 10-12), click here. A middle-1x nm design rule multi-level NAND flash memory cell (M1X-NAND) has been successfully developed for the first time.
1) Quad spacer patterning technology (QSPT) of ArF immersion lithography is used for patterning mid-1x-nm rule wordline (WL). In order to achieve high performance and reliability, several integration technologies are adopted, such as 2) advanced WL air-gap process, 3) floating gate slimming process, and 4) optimized junction formation scheme. And also, by using 5) new N±1 WL Vpass scheme during programming, charge loss and program speed are greatly improved. As a result, mid-1x-nm design rule NAND flash memories has been successfully realized.
The NAND flash memory cell has been scaled down to the 2x [1,2,3] and 2y nm  generations aggressively. As scaling down of a cell size, many serious scaling problems were caused in 2x and 2y nm generation, however they were solved or managed by process, device, and system solutions. For further scaling down beyond 2y nm, we face new scaling limitations such as patterning limitation of ArF immersion spacer patterning technology (SPT), more severe control gate (CG) poly-Si filling problems between floating gates (FGs), and high electric field and charge loss problem between WLs. This paper describes several new advanced processes and operation schemes to overcome these problems, as shown in Table 1. As a result, M1XNAND flash cell is successfully implemented with highly manufacturable integration technologies.
Table 1: Major issues and solution of mid-1x cell technology. M1X-NAND cell process Figure 1 shows the layout of M1X-NAND flash cells. The half pitch of WL is middle-1x nm. The BL contacts are formed staggered arrangement and a string has several dummy WLs. In order to pattern middle-1x nm design rule WLs, QSPT is intensively developed to overcome limitation of ArF immersion SPT. Figure 1: Layout of Mid-1x-nm NAND (M1X-NAND) string with dummy WLs. The half pitch of WL is mid-1x-nm. As shown in Figure 2, first patterns are formed by photolithography and the two times combination of previous formed pattern and spacer are formed final patterns. Figure 2: Schematic diagram of QSPT (Quad Spacer Patterning Technology) key fabrication steps. Two times spacer patterning is used to make mid-1x patterning. The WL critical dimension (CD) of QSPT, which plays a very important role of Vth distribution factor, is precisely controlled less than 1.5% uniformity (see figure 3). QSPT is successfully adopted for mid-1x-nm design rule NAND cell patterning. Figure 3: Location dependence of WL CD variation of QSPT gate patterned NAND string. WL CD is precisely controlled under 1.5% uniformity. Figure 4 shows cross-sectional TEM micrographs of M1X-NAND cell, (a) along WL-direction, and (b) along BL-direction. Figure 4: Cross-section TEM view of the cell, (a) along WL direction, (b) along BL direction. The CGs are well patterned with middle-1x nm half pitch. The floating gate slimming process can achieve the void-free filling of CG poly-Si and wider active area CD, which can obtain large cell current. An electrical depletion in CG poly-Si is greatly suppressed by this void-free process. As a result, BL interference is successfully improved 20% compared with conventional process (see figure 5). The CG CoSi height was selected reasonably to achieve optimized gate shape and decrease WL RC delay for improvement program performance. Figure 5: The simulated results of BL interference with FG slimming scheme. BL interference can be improved to 20%. Cell performance To suppress charge loss (Q-loss) between CG and neighbor FG due to lateral high electric field during program, we have adopted an advanced CoSi-based WL air-gap process that has an air-gap portion above 50% between WLs. As shown in Figure 6(a), the electric field between CG and neighbor FG can be reduced 20% by an advanced CoSi-base WL air-gap. However, reduction electric field by WL airgap is not sufficient to prevent charge loss perfectly because of scale-down issues at mid-1x-nm cell size. So N±1 WL bias control scheme were adopted within WL air-gap. As a result we can also reduce the electric field 15% additionally (see figure 6(a)). Figure 6: (a) Electric field between CG and neighbor FG during programming. (b) 3-D e-field simulation with programmed PV3 neighbor cell. The electric field at point A can be reduced by air-gap and N±1 WL bias control. Then the advanced air-gap and N±1 WL scheme can greatly alleviate charge loss between CG and neighbor FG by decreasing electric field, as shown in Figure 7. Furthermore, as shown Figure 8, PGM speed is improved by N±1 WL scheme, because FG potential of program cell increase by cross coupling effect between WL and neighbor FG. Figure 9 shows the cell coupling ratio with and without air-gap. Cell coupling ratio can be also improved by air-gap due to reducing capacitance of WL direction. Figure 7: Improvement of charge loss with N±1 WL bias control method. Charge loss is greatly decreased to ~300mV by applying Vpass+2V to neighbor N±1 WL. Figure 8: Improvement of PGM speed with N±1 bias control method. Figure 9: Cell coupling ratio as technology shrinkage. Read current reduction is also major issue because of higher bulk doping for suppressing short channel effects in mid-1x-nm cell transistors. A new advanced junction scheme of cell and select transistor is adopted to maximize read current and reduce leakage current in unselected block (see figure 10). Figure 10: Read current with/without select Tr. junction optimization. Read current can be improved by select Tr. junction optimization. Cell Vth Distribution Figure 11 shows cell Vth distributions for the multi-level M1X-NAND cells. The Vth distributions have normal shapes and are well separated to three MLC states. This result confirms that M1X-NAND cell technology can be applied for high density MLC. Figure 11: Three-level programmed Vth distributions of M1X-NAND cells. Vth distributions are well separated to three MLC states. A highly manufacturable mid-1x-nm NAND flash memory (M1X-NAND) has been developed with new integration technologies, such as QSPT, advanced WL air gap process, floating gate slimming process, and optimized junction formation scheme, to overcome scaling limits of mid-1x-nm technology. The excellent device characteristics and reliability are achieved successfully. And also, a new N±1 WL Vpass scheme during programming has been also adopted to overcome WL-to-WL high field issue. Then, we have demonstrated a middle-1x nm-generation NAND flash memory (M1X-NAND) with high performance and reliability. References 1 K. Prall, et al., “25nm 64Gb MLC Technology and Scaling Challenge,” IEEE IEDM Technical Digest, pp. 102-103, 2010. 2. C. Lee, et al., “A Highly Manufacturable Integration Technology for 27nm a and 3bit/cell NAND Flash Memory,” IEEE IEDM Technical Digest, pp. 98-101, 2010. 3. H. Shim, et al., “Highly Reliable 26nm 64Gb MLC E2NAND (Embedded -ECC & Enhanced-efficiency Flash Memory with MSP (Memory Signal Processing) Controller,” VLSI Symp. Technical Digest, pp. 216-217, 2011. 4. K. Lee, et al., “A Highly Manufacturable Integration Technology of 20nm Generation 64Gb Multi-Level NAND Flash Memory,” VLSI Symp. Technical Digest, pp. 70-71, 2011. About the authors This article was contributed by the Flash Device Development & Advanced Process Team, R&D Division, Hynix Semiconductor Inc. The authors include J. Hwang, J. Seo, Y. Lee, S. Park, J. Leem, J. Kim, T. Hong, S. Jeong, K. Lee, H. Heo, H. Lee, P. Jang, K. Park, Myungshik Lee, S. Baik, J. Kim, H. Kkang*, M. Jang*, J. Lee*, G. Cho, J. Lee, B. Lee*, H. Jang, S. Park, J. Kim*, S. Lee, S. Aritome, S. Hong and S. Park