Design Article
Highly scalable vertical gate 3-D NAND
Shih-Hung Chen, Hang-Ting Lue, et al., Macronix International Co.
2/4/2013 4:20 PM EST
Advantages of 3-D VG NAND
There are several advantages of this split-page 3D VG NAND:
(1) Pitch scalability: In this work 3Xnm WL, the smallest in 3D NAND so far, is already demonstrated. Further scaling to 2D NAND limit is likely.
(2) Double-pitch of island-gate SSL, staircase BL contacts, and metal interconnect: These allow larger process window for BL pitch scalability.
(3) Low WL resistance: It is easy to fabricate silicide on top of the WL for WL RC delay reduction. In this work, we use conventional WSix.
(4) Low CSL resistance: Each source contact is directly connected to the local ML1 CSL. A low resistance CSL is critically important for NAND Flash design because the page read demands large current at CSL.
(5) Constant array efficiency at higher stacking layers: As stack number (N) increases, we can simply adjust the BL pad size (shared by 2N pages) but there is no need to expand the array area. Outside the array, only SSL decoder number increases, while WL and BL decoders are kept the same.
MiLC: Minimal Incremental Layer Cost for Staircase Contacts
Staircase contact is a fundamental element for all 3D memories. In a previous proposal [3], multiple trim and etch of photo resist was proposed. However, the PR trim/etch process is not accurate thus not suitable for the tight-pitch staircase contacts. Without new innovation, to make N staircase contacts may require N lithography and etching steps, which would greatly increase the cost. Overlay between lithography steps is also a concern.
We propose a novel method to greatly simplify the staircase BL contacts, as explained in Fig. 4(a). In this work we only use three masks (LA1,2,3, corresponding to 1, 2, 4 poly/oxide etch, respectively) to define a total of 8-layer stack. Sums of binary nodes of 0, 20, 21,…, 2M-1 can generate any integer 0, 1, 2…2M-1. To precisely define the contact position we first use an additional mask to pattern the hard mask and define the contact location precisely. The subsequent LA1, 2, 3 steps carry out the etching of 1, 2, 4 P/O respectively. A layout example is shown in Fig. 4(b). Figure 4(c) shows the fabricated staircase contacts using this novel process. Figures 4(d) and (e) shows the final BL poly plug contacts by this novel “MiLC” process. It shows excellent landing on each memory layer. SiN spacer is used to isolate poly plug and BL pad. Since MiLC allows 2M contacts using only M masks each doubling of layer number only requires one more mask.

Electrical performances of the 8-layer 3D VG NAND array
(1) Island-gate SSL device and Cell Initial I-V: The IdVg curves of our 8-layer island-gate SSL devices are shown in Fig. 5. It shows excellent S.S. behaviors and low leakage, which is important for the NAND string operation. The typical cell initial IdVg curves are shown in Fig. 6.
Figure 5: Typical IdVg curves of the 8-layer island-gate SSL device.

Figure 6: Typical IdVg curves of the 8-layer memory cells.
(2) Programming performance and inhibit of the 8-layer device: Figure 7 shows the ISPP programming and self-boosting inhibit of the 8-layer device. Every memory layer (from PL1 to PL8) can be successfully programmed with memory window greater than 6V, leaving other layers well inhibited.

Click image to enlarge.
There are several advantages of this split-page 3D VG NAND:
(1) Pitch scalability: In this work 3Xnm WL, the smallest in 3D NAND so far, is already demonstrated. Further scaling to 2D NAND limit is likely.
(2) Double-pitch of island-gate SSL, staircase BL contacts, and metal interconnect: These allow larger process window for BL pitch scalability.
(3) Low WL resistance: It is easy to fabricate silicide on top of the WL for WL RC delay reduction. In this work, we use conventional WSix.
(4) Low CSL resistance: Each source contact is directly connected to the local ML1 CSL. A low resistance CSL is critically important for NAND Flash design because the page read demands large current at CSL.
(5) Constant array efficiency at higher stacking layers: As stack number (N) increases, we can simply adjust the BL pad size (shared by 2N pages) but there is no need to expand the array area. Outside the array, only SSL decoder number increases, while WL and BL decoders are kept the same.
MiLC: Minimal Incremental Layer Cost for Staircase Contacts
Staircase contact is a fundamental element for all 3D memories. In a previous proposal [3], multiple trim and etch of photo resist was proposed. However, the PR trim/etch process is not accurate thus not suitable for the tight-pitch staircase contacts. Without new innovation, to make N staircase contacts may require N lithography and etching steps, which would greatly increase the cost. Overlay between lithography steps is also a concern.
We propose a novel method to greatly simplify the staircase BL contacts, as explained in Fig. 4(a). In this work we only use three masks (LA1,2,3, corresponding to 1, 2, 4 poly/oxide etch, respectively) to define a total of 8-layer stack. Sums of binary nodes of 0, 20, 21,…, 2M-1 can generate any integer 0, 1, 2…2M-1. To precisely define the contact position we first use an additional mask to pattern the hard mask and define the contact location precisely. The subsequent LA1, 2, 3 steps carry out the etching of 1, 2, 4 P/O respectively. A layout example is shown in Fig. 4(b). Figure 4(c) shows the fabricated staircase contacts using this novel process. Figures 4(d) and (e) shows the final BL poly plug contacts by this novel “MiLC” process. It shows excellent landing on each memory layer. SiN spacer is used to isolate poly plug and BL pad. Since MiLC allows 2M contacts using only M masks each doubling of layer number only requires one more mask.

Click image to enlarge.
Figure 4: (a) The MiLC concept. It uses binary sum of a few etching steps, each corresponds to 2n in depth, to carry out multiple staircase contacts with minimized incremental layer cost. (b) The layout example. Three masks: LA1 for 1-P/O ETCH, LA2 for 2-P/O ETCH, and LA3 for 4 P/O ETCH are used to carry out 8-layer contacts. (c) Before poly plug, a sidewall lateral recess is carried out to enhance the isolation between poly plugs and BL pad. (d) The final TEM cross-sectional view. (e) Zoom-in views of MiLC.
Electrical performances of the 8-layer 3D VG NAND array
(1) Island-gate SSL device and Cell Initial I-V: The IdVg curves of our 8-layer island-gate SSL devices are shown in Fig. 5. It shows excellent S.S. behaviors and low leakage, which is important for the NAND string operation. The typical cell initial IdVg curves are shown in Fig. 6.
Figure 5: Typical IdVg curves of the 8-layer island-gate SSL device.

Figure 6: Typical IdVg curves of the 8-layer memory cells.
(2) Programming performance and inhibit of the 8-layer device: Figure 7 shows the ISPP programming and self-boosting inhibit of the 8-layer device. Every memory layer (from PL1 to PL8) can be successfully programmed with memory window greater than 6V, leaving other layers well inhibited.

Click image to enlarge.
Figure 7: Typical programming and self-boosting inhibit characteristics of the 8-layer memory cells measured in our 3D VG array. Within one page (SSL) and a selected center WL, we program only one layer, leaving other memory cells inhibited. (a) Program PL8 only. (b) Program PL7 only. (c) Program PL6 only. (d) Program PL5 only. (e) Program PL4 only. (f) Program PL3 only. (g) Program PL2 only. (h) Program PL1 only. All 8-layer devices are successfully programmed. The self-boosting program inhibit is also successful.
Navigate to related information


greenpattern
2/5/2013 11:58 PM EST
These 3D-NAND structures essentially use fork-shaped capacitors - that needs to be taken into account.
Sign in to Reply
resistion
2/6/2013 9:16 AM EST
Fork shape?
Sign in to Reply
greenpattern
2/6/2013 11:38 AM EST
The WL's and BL's are not individual lines but linked within the same plane, forming forks. The BL fork actually touches source line, not actual fork shape sorry. Voltage over entire plane being same, the capacitive coupling to neighboring plane is serious risk.
Sign in to Reply
resistion
2/10/2013 11:33 AM EST
So there is larger capacitance between BL's or WL's than in the gate stack itself?
Sign in to Reply