High-tilt implant and diffusion-less activation for lateral graded S/D engineering
06/01/2003
Overview
As the industry continues to scale devices to meet the various system-on-a-chip applications in the future, several types of transistor designs from planar and vertical single-gate, double-gate, and multigate devices are emerging, as are several options in silicon starting material (bulk CZ, epi, blanket SOI, and selective/patterned SOI wafers) [1–7]. Although planar single-gate device scaling has been demonstrated down to a gate length (Lg) of 6nm, the end of the roadmap for planar single-gate CMOS seems to be drawing nearer as the industry increases research activities in double-gate and multigate CMOS devices [8]. Multiple device roadmaps that depend on specific applications are also emerging and both logic and memory devices will migrate from planar structures to vertical structures.
One of the critical areas in future transistor design scaling for improved device performance will be lateral source/drain (S/D) dopant engineering, including lateral abruptness, gate overlap control, and lateral dopant activation. Recent advances using high-tilt implantation and diffusion-less activation starting at the 65nm node and extendable to at least the 23nm node (Lg = 13nm) based on the roadmap (see the table) will be described [9].
Single-gate structure
There is no fundamental change in the design of the planar single-gate (SG) transistor to about 2015. However, changes for improved device performance will be made for its scaling and extendibility. The current oxynitride gate dielectric material will first evolve to a medium-k dielectric of 12 (HfSiON, HfO2/.Al2O3, etc.) at the 65nm node and then to a high-k of 25 (HfO) by the 45nm node. High-temperature diffusion-less activation (flash/RTA [rapid thermal anneal] or submelt laser annealing) can be used with medium-k gate dielectrics, but once high-k gate dielectrics are implemented, only low-temperature, diffusion-less activation, solid phase epitaxy (SPE) can be used to maintain their amorphous phase and high-k integrity (Fig. 1) [9–11]. The boron solid solubility (Bss) dopant activation values based on a box profile shape were calculated (atoms/cm3) and identified (see the table and Fig. 1). Possible design evolutions of the transistor channel and S/D regions from the 90nm node through the 32nm node were collected from the top 10 logic companies worldwide [12; Fig. 2].
300mm silicon wafer options
For improved silicon material quality at 300mm, a majority of IC manufacturers have switched to epi wafers to maximize yield. Growing an epitaxial silicon layer on top of bulk Cz wafers eliminates grown-in defects such as crystal-originated precipitate [13, 14]. For the same reason, silicon-on-insulator (SOI) wafers are also using a top epitaxially grown silicon layer for both the SIMOX and wafer-bonding manufacturing methods in order to achieve the highest SOI silicon layer quality [15].
 |
null
One of the main challenges for thin SOI wafer manufacturing is scaling the film thickness with each technology node while maintaining tight uniformities. At the 180nm node, the SOI film thickness for partially depleted (PD) SOI devices was 160–200nm. For the 130nm node, the SOI film thickness was reduced to a range between 79–120nm and for the 90nm node, the film thickness range is expected to be between 30–60nm. For this reason, PD/SOI film thickness is approximately equal to Lg [2, 3, 5, 6]. With thinner SOI films, tighter control on thickness uniformity is critical since it directly defines the device channel region and variations will directly impact Vt control and ring oscillator delay as reported by Shahidi [6]. For fully depleted (FD) SOI devices, the SOI film thickness is targeted to be about 0.2–0.3¥Lg, so at the 65nm node with Lg = 35nm, the SOI film is expected to be only 7–10nm thick [2, 3]. Very thin SOI film thickness uniformity and control for these fully depleted devices will be a major challenge, especially for bonded SOI wafer manufacturers in the future. Also, as the SOI top silicon film gets thinner, S/D parasitic resistance goes up and contacting silicide becomes a problem requiring the use of elevated S/D structures.
For embedded memory and other system-on-a-chip (SOC) applications, logic devices on SOI material and memory devices on bulk material may be preferred, creating a new classification of SOI wafers called selective/patterned SOI wafers. Two different approaches to realize logic devices in SOI and DRAM trench capacitor memory devices in bulk (non-SOI) have been recently reported. Researchers at IBM have used the SIMOX method with a patterned mask for the oxygen implant, resulting in regions on a bulk wafer with patterned SOI structures [5]. At Toshiba, researchers used the bonded SOI wafer approach with patterned etch back through selected SOI regions down to the bulk wafer, followed by selective epitaxial growth in these non-SOI regions, forming selective SOI [16]. Another SOC embedded device application is the bipolar RF device, where the vertical SiGe bipolar device can be on bulk (non-SOI) or SOI material [6].
Channel and S/D engineering
Starting at the 90nm node, higher-dose HALOs (super HALO) will be used for short channel effect (SCE) control, reducing Vt roll-off [12; Fig. 2]. As Lg scales, however, the HALOs will become overlapped (Fig. 2), raising the channel-doping level into the mid- to upper E18/cm3 levels [1]. Impurity scattering occurs as the channel-doping level becomes >2E18/cm3, thereby degrading channel mobility and leading to the adoption of the strained-Si channel [17, 18]. The best strained-Si on SiGe epilayer has about 105 defects/cm2, so compromises have been made to achieve strained-Si by combination of other techniques such as 1) localized Ge implantation of a few percent; 2) strain-induced CVD layer deposition; and 3) strain-induced STI formation. The use of strained Si technology can be avoided with the lateral graded S/D terraced (quad/multimode implant) or wedged (rotation-mode implant) structure improving lateral abruptness, gate overlap control, SCE, and Rext, thereby reducing the HALO dose [4].
 Figure 1. Diffusion-less activation roadmap. Bss is based on box profile boron solid solubility limit. |
null
Current transistor designs use separate shallow source/drain extension (SDE) junction and deep S/D contacting junction implants and masks. After efforts to maximize device drive current, Ghani et al. reported on the criticality of the optimization of SDE vertical junction depth (Xj) and lateral junction depth (Yj) called gate overlap/under diffusion and lateral abruptness [17]. To control the SDE gate overlap precisely, off-set spacers were introduced to compensate for excessive lateral diffusion for both bulk and SOI devices becoming standard at the 90nm node [19–21]. The updated roadmap (see the table) has the new target values for Xj, Yj, lateral abruptness, and box profile activated dopant concentration for the most advanced logic devices taken from several end-user inputs.
For high-performance (HP) logic devices using oxynitride gate dielectric with poly electrode, high-temperature dopant activation annealing can be used for the SDE, deep S/D, and poly-gate-doping activation and diffusion (>10nm). Starting at the 65nm node, however, the ultra-shallow junction (USJ) requirements for SDE necessitate diffusion-less (<3.0nm) dopant activation annealing at either high temperatures or low temperatures, while the deep S/D and poly-gate doping will still require some dopant diffusion (>10nm). Therefore, from a process integration point of view, several companies will adopt the disposable spacer process flow also called reverse S/D for SDE. This adoption will allow the use of traditional high-temperature spike/RTA annealing with >10nm dopant diffusion for the deep S/D and gate poly doping. Following this will be either high-temperature flash (<4nm dopant diffusion), submelt laser annealing (<1nm dopant diffusion), or low-temperature SPE (solid phase epitaxy) annealing (no dopant diffusion) for diffusion-less dopant activation to achieve shallow and abrupt SDE junctions with gate overlap control (Fig. 2).
 Figure 2. HALO overlapping with Lg scaling [1]. |
null
For low-power logic devices using high-k gate dielectrics and metal electrodes starting at the 45nm node, low-temperature dopant activation techniques will be required to prevent crystallization of the amorphous high-k gate material. Fortunately, the acceptable levels of junction leakage current for HP and low-operating-power (LOP) logic devices are within the acceptable range observed with low-temperature SPE diffusion-less annealing. Low-standby-power (LSTP) logic devices may require a combination of low-temperature SPE annealing first, followed by higher-temperature RTA annealing to further improve junction leakage [9].
Further improvements in device performance can also be achieved with single-S/D, the merging of shallow SDE junction and deep S/D contact junction into a single junction as reported by Ozterk et al. [22]. Single-S/D structures will require the use of elevated S/D contacting structures in order to achieve low-S/D parasitic resistance and good salicide contacting just as PD/SOI devices at 90nm and beyond with thin SOI <40nm thick now require undoped elevated S/D structures. For this reason, single-S/D structures will first be implemented in SOI technology, while bulk non-SOI devices will still require deep S/D to maintain low-S/D parasitic resistance and good salicide contacting. Some companies have proposed a third, deeper contacting junction for bulk CMOS devices [23, 24]. Once bulk CMOS devices switch to elevated S/D to maintain good junction contacting, as the contacting S/D junction scales below 40nm at the 45nm node, they too could merge the deep and shallow S/D junctions, forming single-S/D.
A simple way to transition from separate shallow SDE and deep S/D junction structures to a single-S/D structure is by using multiple, high-tilt, quad-mode implantation through the off-set spacer to form lateral graded SDE (Lg-SDE) or single S/D (Lg-SS/D) called PoGI (post-gate implant) [12; Fig. 3]. Diffusion-less activation simulation results for sub-65nm node devices (Lg = 25nm, height = 65nm, and Tαx = 0.8nm) are shown in Figs. 3 and 4 for a Taurus Monte Carlo implant model developed by Synopsys. An arsenic 3keV/1E15/cm2 implant was used for the nSDE. Results for the traditional zero tilt nSDE showed no gate overlap, while a 30°, high-tilt, PoGI quad-mode implant (0, 90, 180, and 270° wafer twist) for an arsenic 3keV/1E15/cm2 implant is shown in Fig. 3, realizing 8.5nm of gate overlap (encroachment). Geometrically, sin 30° = 0.5, so Yj = 0.5Xj and this is what the simulations also show. Also, the ratio of Yj/Xj is approximately 0.5 for the 130nm, 90nm, and 65nm nodes (see the table) and increases to 0.6 (sin 35°) for the 45nm and 32nm nodes. So a tilt of 30–35° seems optimum with diffusion-less activation for gate overlap control.
Both zero and 30° high-tilt implantation results satisfy the 65nm node for nSDE Xj = 10–17nm depth and lateral abruptness of <2.8nm/decade. However, the Yj gate overlap target of 8nm could not be achieved for zero tilt (0nm overlap), while for the 30° high-tilt case Yj = 8.5nm using a 5nm off-set spacer results in a net gate overlap/encroachment under the gate edge of 3.5nm, thus enabling the use of a thinner off-set spacer. With the traditional zero-tilt implant, the peak arsenic dopant concentration at the gate edge is zero, and under the sidewall spacer it is 5E20/cm3. With the LG-SDE terraced structure, the peak arsenic dopant concentration at the gate edge is 5E20/cm3, and under the sidewall spacer it is 1.3E21/cm3, resulting in a lower value of Rext and the length of SDE is now controlled by the gate stack height shadowing effects (sin X) and not the sidewall spacer thickness variations.
As the industry moves toward diffusion-less activation (65nm node and beyond) for shallow junction formation, high-tilt directional implantation in the lateral dimension will become desirable to precisely control gate overlap (encroachment); this is now on the roadmap of several companies. All diffusion-less activation processes (high-temperature flash/RTA and submelt laser or low-temperature SPE annealing) that use some form of pre-amorphizing implant (PAI) for the SDE gate overlap precision control and optimization of lateral dopant activation, will also require high tilt for the PAI implantation step.
A simulation comparison between a 10keV/5E14/cm2 Ge-PAI 0° tilt and 30° tilt for a B11 = 500eV/1E15/cm2, 30°-tilt Lg-SDE was performed (Fig. 4). Outlined in black is the simulated amorphous/crystalline silicon interface — vertical and lateral. No lateral placement under the gate sidewall for zero tilt was observed, while a 7.1nm (Yj/Xj = 0.5) lateral placement under the gate with the 30° tilt was achieved as expected. High-tilt dopant and PAI implantation will therefore allow precise optimization and engineering of lateral dopant activation for gate overlap control, especially for diffusion-less activation (high- or low-temperature) processing starting at the 65nm node. P. Chidambaram et al. recently reported on improved 50nm pMOS devices using tilted PAI implantation [25]. Reduction in overlap capacitance of 8%, and a 3% improvement in drive current were reported. Additionally, improvements were found in lateral abruptness, SCE, Vt roll-off, as well as decreased channel doping.
Various options for the LG-SS/D structure when using PD/SOI devices with elevated S/D were performed (Fig. 5). Rotating the wafer for multiple twist implants, i.e., 4–8 rotations for a given high-tilt implant, will create a double (case B for a 45° quad twist), triple (case A for a 0° quad twist), or penta (cases A and B for both 0° and 45° quad twist) terraced LG-SS/D structure. It is clear that additional/continuous rotation will lead to a wedged structure. Since the tilted implant is done through the off-set spacer, the integration of this process can result in two mask savings with elevated S/D in SOI or bulk technology. If the bulk technology structure requires deep S/D for salicide contacting, then a disposable spacer process flow could be used to save two mask levels.
The potential device performance benefits of the LG-SDE and Lg-SS/D terraced or wedged structures are: 1) improved gate overlap control and SCE due to shallower junction (Xj), more gate overlap (Yj), and good lateral abruptness; 2) reduced HALO dose to avoid channel mobility degradation and delay the need for strained-Si channel technology; 3) reduced gate overlap capacitance and junction capacitance; and 4) improved Idsat and increased ring oscillator speed due to lower Rext. If elevated S/D structures are used, then a selective poly deposition process is preferred over a selective epi deposition process to avoid issues with sidewall-faceting control and interface surface-cleaning sensitivity. This can be achieved using controlled interface cleaning with selective HF vapor etching, for example [26, 27].
 Figure 3. Taurus Monte Carlo implant simulation for 30° quad-mode arsenic 3keV/1E15 implant through off-set spacer. |
null
 Figure 4. High-tilt 30° vs. zero-tilt Ge-PAI implantation for pSDE gate overlap control and optimization of lateral dopant activation when using diffusion-less activation. |
null
Elevated S/D and contact salicide
Undoped elevated S/D structures for improved salicide contacting will first be used for thin (<40nm thick) PD/SOI devices at the 90nm node to maintain good salicide S/D contacting. For sub-90nm PD/SOI, the top silicon layer thickness (24–30nm) is approaching the depth of the SDE (15–24nm), and the SDE and deep S/D junctions are merging quickly into single-S/D, making Lg-SS/D structures ideal for lower S/D tip resistance for thin PD/SOI devices beginning at the sub-65nm node. As the deep S/D junction on bulk CMOS devices scales below 40nm junction depth, the industry will also adopt elevated S/D at Lg between 20–40nm. The selective silicon (epi or poly) deposited structure does not have to be doped, provided the silicide contact metallization process step completely consumes the planar facet-free, elevated silicon structure [28–30].
The switch from CoSi2 to NiSi2 salicide for lower-S/D contact resistance and poly-gate resistance will start at the 90nm node for compatibility with SiGe material used in strained Si channel technology, lower contact, and gate/poly resistance and lower-temperature processing. To achieve good contacting, a certain amount of silicon must be consumed. This becomes problematic with SOI devices, where the S/D parasitic resistance increases with thinner SOI layers. The thin junction layer under the silicide and small interface contact area results in higher resistance and, therefore, the need for elevated S/D [31].
 Figure 5. Lg-SS/D terraced structure dependence on PD/SOI wafer twist during quad-mode implant with elevated S/D. |
null
Double/multigate CMOS
With expected difficulties in the scaling of SG FD/SOI wafers and devices, the industry is rapidly looking at double-gate (DG) and multigate (MG) FD/SOI devices for possible introduction at the 32nm node for Lg10–20nm. This would ease SOI wafer manufacturing, since the SOI thickness will be 0.7–1¥Lg compared to 0.2–0.3¥Lg for SG FD/SOI. Other benefits include an undoped-to-lightly-doped channel (thus avoiding mobility degradation and easing concern over the random dopant fluctuation effects in the channel region), better SCE control, and higher drive current [2, 3]. Also, lateral graded S/D dopant engineering and elevated S/D contact formation technique will be used for both DG and MG novel devices as described earlier for the SG device architectures.
 Figure 6. Single- and multi-FIN-FET double-gate device [34]. |
null
Planar (top/bottom) DG CMOS
One of the most difficult challenges, and therefore critical requirements, in the planar DG structure is the alignment of the top gate to the bottom gate [12; Fig. 1]. The simplest method is using wafer-bonding techniques, although gate alignment is most difficult. Another approach is to use selective epitaxy with lateral growth through a tunnel to form the source, and then the channel (tunnel), and last, the drain. For this structure, no SOI starting wafer is needed, but this is a complicated structure to achieve.
FIN-FET DG and MG CMOS
The FIN-FET DG structure is the easiest to fabricate and currently the most popular DG structure found in the literature. For the FIN-FET design, the SOI silicon film thickness determines the minimum gate width (Wg) dimension, so thicker SOI layers are required. With this structure, high-tilt implantation up to 45° for SDE and S/D doping prior to elevated S/D for improved salicide contacting will be needed as reported by Kedzierski et al. [32]. Additional flexibility for designing asymmetrical n+/p+ poly gates by high-tilt poly-gate implant doping is another option in transistor design for undoped channels. If Wg is scaled down to equal Lg, then a triple-/multiple-gate structure can easily be realized [33]. Again, multiple high-tilt implantation for LG-SDE and LG-SS/D structures can be used with the various FIN-FET device structures (Fig. 6) [34]. To create devices with various gate widths, multi-FIN structures will be required [34]. This 3-D poly-gate structure formed around multiple FIN structures may require conformal doping by plasma implantation (Fig. 6).
Vertical pillar DG CMOS
For the pillar DG design, also called vertical gate replacement (VGR) transistor, an SOI wafer is also not necessary and the S/D, SDE, and channel regions are selectively epitaxially grown and doped. If the gate structure goes completely around the pillar or planar DG structure, the result is a quadruple-gate, called gate-all-around or surround-gate [12; Fig. 1].
Conclusion
Multiple types of transistor designs and silicon wafers will be used as the industry continues to scale devices to meet the SOC applications of the future, ranging from planar SG and vertical SG structures, to vertical DG and MG options. Various silicon wafers will also be used, including selective/patterned SOI wafers. One of the key areas for extending SG and advancing DG transistor performance will be lateral graded S/D engineering, realized with high-tilt implantation and diffusion-less activation starting at the 65nm node.
John O. Borland, Varian Semiconductor Equipment Associates, Newburyport, Massachusetts; Victor Moroz, Synopsys, Mountain View, California; Howard Wang, TSMC, Hsinshu,
Taiwan, R.O.C.; Witek Maszara, Advanced Micro Devices, Sunnyvale, California;
Hiroshi Iwai, Tokyo Institute of Technology, Yokohama, Japan
References
1. H. Iwai, VSEA vTech seminar pres. mat., San Francisco, CA, July 22, 2002.
2. W. Maszara, MRS Symp. Proc., Vol. 686, p. 59, 2001.
3. W. Maszara, VSEA vTech seminar pres. mat., July 22, 2002.
4. J. Borland, VSEA vTech seminar pres. mat., San Francisco, CA, July 22, 2002.
5. D. Sadana, VSEA vTech seminar pres. mat., San Francisco, CA, July 22, 2002.
6. G. Shahidi, VSEA vTech seminar pres. mat., San Francisco, CA, July 22, 2002.
7. R. Divakaruni, G. Bronner, The Electrochemical Society, PV2001-2, March 2001.
8. B. Doris et al., IEDM-2002, section 10.6, p. 267, Dec. 2002.
9. J. Borland, extended abstract of the International Workshop on Junction Technology, p. 85, Dec. 2002.
10. M. Niwa, short course pres. mat., IEEE 2000 Symp. on VLSI Tech., June 2000.
11. J. Borland, T. Matsuda, S. Sakamoto, Solid State Technology, Vol. 45, No. 6, p. 83, June 2002.
12. J. Borland, H. Iwai, W. Maszara, H. Wang, Solid State Technology, Vol. 46, No. 3, p. 26, March 2003.
13. J. Borland, Semiconductor International, part 1, p. 144, April 1989.
14. J. Borland, Semiconductor International, part 2, p. 154, May 1989.
15. J. Borland, Semiconductor International, p. 70, April 2001.
16. T. Yamada et al., VLSI Symposium 2002, section 12.1, June 2002.
17. T. Ghani et al., IEEE 2000 VLSI Symp., section 18.1, p. 174, June 2000.
18. S. Thompson et al., IEDM-2002, section 3.2, p. 61, Dec. 2002.
19. S. Thompson, short course pres. mat. at the IEDM-1999.
20. S. Narasimha et al., IEEE IEDM-2001, section 29.2, p. 625, Dec. 2001.
21. Y. Inoue, VSEA vTech seminar pres. mat., July 22, 2002.
22. M. Ozterk et al., The Electrochemical Society, May 2002 meeting, to be published.
23. J. Check et al., The Electrochemical Society, PV99-18, p. 275.
24. C. Wang, M. Chen, Mosel Vitelic, US patent Oct. 27, 1998, #5,827,747.
25. P. Chidambaram et al., extended abstract #893, Spring Electrochemical Society meeting, Paris, France, April 2003.
26. J. Borland, Univ. of Calif. at Berkeley extension class on polysilicon lecture notes 1993–2000.
27. J. de Larios et al., SSDM-93, p. 140, Sept. 1993.
28. J. Borland, IEDM-97, section 2.1, p. 12, Dec. 1997.
29. C. Wei et al., IEEE VMIC-1989, p. 136.
30. C. Wei et al., The Electrochemical Society, PV 89-9, p. 637, 1989.
31. D. Hisamoto, IEEE IEDM-2001, section 19.3, p. 429, Dec. 2001.
32. J. Kedzierski et al., IEEE IEDM-2001, section 19.5, p. 437, Dec. 2001.
33. R. Chau et al., SSDM-2002, p. 68, Sept. 2002.
34. Y. Choi et al., IEEE IEDM-2001, section 19.1, p. 421, Dec. 2001.
For further information, contact John O. Borland, Varian Semiconductor Equipment Associates, 4 Stanley Tucker Dr., Newburyport, MA 01950; ph 978/463-5070, fax 978/462-0210, [email protected].