Professional Documents
Culture Documents
Design Technology of Stacked Type DTMOS: Yu Hiroshima, Takahiro Kodama, and Shigeyoshi Watanabe
Design Technology of Stacked Type DTMOS: Yu Hiroshima, Takahiro Kodama, and Shigeyoshi Watanabe
Design Technology of Stacked Type DTMOS: Yu Hiroshima, Takahiro Kodama, and Shigeyoshi Watanabe
7, 2014
Translated from Denki Gakkai Ronbunshi, Vol. 132-C, No. 12, December 2012, pp. 1927–1933
55
Fig. 1. Structure of staked DTMOS.
56
FinFET DTMOS and stacked DTMOS are assumed equal
to design rule F. The sidewall channel width Dconv of
the conventional FinFET DTMOS is set to 2.5 F, and the
sidewall channel width of the stacked DTMOS is set to
5 F. The 70 nm design rule is assumed, and the impurity
concentration in the substrate portion is 1017 cm−3 . The
substrate resistance RS of the conventional FinFET DT-
MOS calculated using this values is about 35 KΩ. The
stray capacitance CS of the substrate calculated from the
gate capacitance and junction capacitance is about 0.5 fF.
Fig. 3. Parameters related to delay time of substrate. Thus, the substrate’s time constant, defined as the product
of RS and CS, is about 18 ps. This value is greater than the
12 ps delay time of the basic NAND circuit of a GHz-class
characteristics, and a PMOS (100) substrate with a thick- microprocessor [12]. That is, in terms of the substrate’s
ness of 5 F = 70 nm × 5 = 350 nm can be realized by delay time, the conventional FinFET DTMOS is suitable
adjusting the acceleration voltage and dose of implanted for microprocessors operated at about 100 MHz, but opera-
oxygen ions (in Ref. [8], a PMOS (100) substrate about tion in GHz band is problematic. In addition, the transistor
100 nm thick was obtained at an acceleration voltage of threshold voltage may be widely distributed because of the
130 keV and a dose of 4.5× 1017 cm−2 ). After that, an difference in distance from the gate-substrate connection to
insulating oxide with a thickness of about 70 nm (F) is the substrate. In contrast, in the stacked DTMOS, the gate
deposited (in Ref. [10], an oxide film of 30 nm to 100 nm and substrate are connected on the sidewall, and thus the
was formed). substrate resistance RS can be reduced to less than 10%
Then, an NMOS (110) substrate with different ori- of that in the conventional FinFET DTMOS, that is, to
entation is formed on top by bonding (a). The (110) sub- 35 × (1/2.5) × (1/5) = 2.8 KΩ. Accordingly, the delay time
strate is employed because the (100) plane, with the high- too can be reduced to 1.4 ps, which is suitable for GHz
est electron mobility, is used for electrical conduction of operation. The distance from the gate-substrate connection
the NMOS FinFET formed perpendicular to the substrate. to the substrate is constant in the stacked DTMOS, and
The desired thickness of the NMOS (110) substrate can be hence there is no concern about wide distribution of the
obtained by adjusting the acceleration voltage and dose of transistor threshold voltage.
the implanted oxygen ions (in Ref. [10], the (110) substrate Next, we explain how the PMOS driving performance
was bonded to the (100) substrate formed by SIMOX via can be improved at the LSI level by taking account of the
an insulating film). After that, two lateral surfaces of the substrate plane orientation during the fabrication of the
PMOS substrate and two lateral surfaces and the top surface stacked structure. The substrate plane orientation depen-
of NMOS substrate are oxidized (b). Then, the oxide film dence of transistor’s mobility was previously investigated
is removed from one lateral surface of the NMOS substrate to enhance the current drivability of MOS transistors. In
and the PMOS substrate (c). In the case of conventional planar transistors where current flows in the direction of
FinFET reported in Ref. [11], a single-sided film removal the crystal plane, the current drivability of both NMOS
process is used to obtain different thicknesses of the gate and PMOS is known to be maximized when NMOS is
insulating films on the two sides of the gate. Thus, a single- formed on (the 100) substrate and PMOS is formed on the
sided film removal process seems industrially feasible. Fi- (110) substrate [13, 14]. In the common case in which the
nally, the gate electrode is formed (d). Thus, the stacked planar NMOS and PMOS structures of LSIs are formed on
DTMOS shown in Fig. 1 can be implemented with relative substrates with the same plane orientation, (100) substrates
ease. are employed to seat NMOS, which higher mobility than
PMOS, thus sacrificing the current drivability of the PMOS.
In this approach, no additional pattern area is required
3. Acceleration of Operation by Introduction of but the current supply capacity of the PMOS drops by
Stacked DTMOS about 40% compared to the optimized mobility conditions
[15–17].
In this section, we estimate how much the operation In FinFET, where current flows in the direction per-
speed can be improved by the introduction of stacked DT- pendicular to the crystal plane, the current drivability of
MOS. both NMOS and PMOS can be maximized when the NMOS
First, we estimate the time constant from the gate- and PMOS are arranged at an angle of 45◦ (135◦ ) on the
substrate connection to the substrate. The parameters used same (100) substrate, as shown on the left of Fig. 4.
for estimation are shown in Fig. 3. Both the silicon pil- That is, by using this method, maximum current
lar width W and the gate length L in the conventional drivability of both NMOS and PMOS is implemented on
57
Table 1. Relations between selection of substrate
orientation/placement of FET and delay time/pattern area
58
Table 2. Design rule for pattern design
59
Fig. 8. Pattern area comparison for system LSI for
communication: (a) planar, (b) planar DTMOS, (c)
FinFET DTMOS, (d) stack DTMOS.
Fig. 7. Two-input NAND/NOR: (a) circuit diagram, (b)
planar layout, (c) planar DTMOS layout, (d) FinFET
DTMOS layout, (e) stacked DTMOS layout.
60
and 5 F of communication LSIs. Analysis showed that the duction of the pattern area. The gate and substrate are con-
pattern area is minimized at 7.5 F and 15 F, respectively. In nected using the whole area of the silicon pillar sidewall,
this case, too, stacked DTMOS offered a greater reduction and as a result, the substrate’s delay time can be reduced
of the pattern area than FinFET DTMOS. below 1/10 compared to the conventional FinFET DTMOS.
In addition, the NMOS is stacked on the (110) plane on
top of PMOS formed on the (100) plane, and as a result,
6. Future Prospects
high-speed operation can be achieved by using the optimal
mobilities of both the NMOS and PMOS without requiring
In this investigation, we examined the operation
additional pattern area. We applied the stacked DTMOS to
speed, determined by the substrate’s time constant and the
inverter, NAND, and other simple logic circuits as well as to
mobility of NMOS and PMOS at a 70-mm design rule.
a full adder, a communication LSI, and a DRAM buffer cir-
We now compare the operation speed of the conventional
cuit. We found that the pattern area was reduced compared
FinFET DTMOS and the proposed stacked DTMOS when
to the conventional FinFET DTMOS, to 44% to 86% for
the design rule is reduced below 70 nm.
the inverter, NAND, and other simple logic circuits, 89%
DTMOS has a partial depletion structure and it is
for the full adder, 84% for the communication LSI, and
difficult to reduce the impurity concentration in the sub-
58% for the DRAM buffer circuit. The proposed stacked
strate on miniaturization. Thus, we assumed an impurity
DTMOS is a promising solution for future implementation
concentration of 1017 cm−3 , as with the 70 nm design rule.
of microprocessors for portable devices operated in the
Thus, the ratio of the substrate time constants of the FinFET
GHz band and other advanced system LSIs.
structure and the stacked structure can be expressed as (Fin-
FET silicon pillar width)/(FinFET silicon pillar height) ×
(stacked silicon pillar width)/(stacked silicon pillar height).
In terms of the design rule, this value is (F/2.5F) × (F/5F)
REFERENCES
= 0.08, and thus is constant regardless of the design rule.
Thus, we can assume that the superiority of the stacked
1. Assaderaghi F, et al. Dynamic threshold-voltage
structure over the FinFET structure remains unchanged
MOSFET (DTMOS) for ultra-low voltage VLSI. IEEE
even when a device is miniaturized. If, with further device
Trans Electron Dev 1997;44(3):414–422.
miniaturization, a greater aspect ratio of the silicon pillar
2. Kakimoto S, et al. CMOS device technology for ul-
(height/width ratio) were required for implementation of
tra low power LSIs. Sharp Tech J 2001;79:16–21. (in
transistors with higher density and better performance [11],
Japanese)
then the substrate’s time constant ratio would be (F/2.5kF)
3. Hiroshima Y, Watanabe S. Proposal of a FinFET type
× (F/5kF) = 0.08k2 , where k denotes rate of increase of
DTMOS. Trans IEICE 2009;J92-C(11):742–743. (in
the aspect ratio with respect to 70-nm design; that is, the
Japanese)
stacked structure would become even more advantageous.
4. Jain S, et al. A 280mV-to-1.2V wide-operating IA-32
The mobility of NMOS and PMOS decreases as the
processor in 32nm CMOS. ISSCC Dig. Tech. Papers,
electric field strength increases with smaller design rules;
2012.
however, there are hardly any data about significant differ-
5. Intel 22nm 3-D Tri-Gate Transistor Technology:
ences in the decrease rate depending on the plane orienta-
http://download.intel.com/newsroom/kits/22nm/pdfs/
tion. In addition, it has been reported that when the width of
22nm-Announcement Presentation.pdf.
the silicon pillars decreases to 8 nm, the mobility changes
6. Davnaraju S, et al. A 22nm IA multi-CPU and GPU
due to quantum effects, the surface roughness of pillar, and
system on chip, ISSCC Dig. Tech. Papers, 2012.
other factors [15]. However, these effects are not likely to
7. Horita K, et al. Layout-independent transistor
occur at design rule of 10 nm and larger, as assumed in this
with stress-controlled and highly manufacturable
study.
shallow trench isolation process. J Appl Phys
Thus, we can expect that the ratio of the substrate
2007;46(4B):2079–2083.
time constants between the FinFET structure and the
8. Wu X, et al. A three-dimensional stacked Fin-CMOS
stacked structure, as well as the ratio the of NMOS and
technology for high-density ULSI circuits. IEEE Trans
PMOS mobilities between different plane orientations, is
Electron Dev 2005;52(9). (in Japanese)
unlikely to depend on the design rule.
9. Hiroshima Y, Watanabe S. New design technology of
independent-gate controlled stacked type 3D transistor
7. Conclusions for system LSI. Trans IEICE 2009;J92-C(3):94–103.
(in Japanese)
We have proposed a new stacked DTMOS that com- 10. US Patent: Structure and method of manufacturing a
bines the FinFET DTMOS, with high-speed and low-power FinFET device having stacked fins. US 2005/0339242
consumption, and stacked transistors, with considerable re- A1, 2005.
61
11. Liu YX, et al. Advanced FinFET CMOS technology: formulas. IEEE J Solid-State Circuits 1990;25(4):584–
TiN-gate, fin-height control and asymmetric gate in- 594.
sulator thickness 4T-FinFETs. IEDM Dig of Tech Pa- 20. Ishikuro H, Hamada M, Agawa K, Kousai S,
pers, 2004. Kobayashi H, Nguyen D, Hatori F. A single-chip
12. International Technology Roadmap for semiconduc- CMOS Bluetooth transceiver with 1.5 MHz IF and
tors. 2003 edition. direct modulation transmitter, ISSCC Dig Tech Papers
13. Sato T, et al. Mobility anisotropy of electrons in inver- 2003:68–69. (in Japanese)
sion layers on oxidized silicon surfaces. Phys Rev B, 21. Watanabe S. Design methodology for system LSI
Condens Matter 1971;4:1950–1960. with TIS (Trench-Isolated transistor using Sidewall
14. Kinugawa M, et al. Effects of silicon surface orien- gate). Trans IEICE 2005;J88-C(12):1208–1218. (in
tation on submicron CMOS devices. IEDM Tech Dig Japanese)
1985:581–584. 22. Oowaki Y, Tsuchida K, Watanabe Y, Takashima
15. Poijak M, et al. Modeling study on carrier mo- D, Ohta M, Nakano H, Watanabe S, Nitayama
bility in ultra-thin body FinFETs with circuit-level A, Horiguchi F, Ohuchi K, Masuoka F. A 33-
implications. Solid-State Electron 2011;65–66:130– ns 64Mb DRAM. IEEE J Solid-State Circuits
138. 1991;26(11):1498–1505.
16. Tsutsui G, et al. IEEE Electron Dev Lett 2005;26:836– 23. Watanabe S. New design method for tapered buffer cir-
838. cuit using TIS (Trench-Isolated transistor using Side-
17. Kobayashi S, et al. Tech Dig IEDM 2007:707–710. wall gate) and its application to high density DRAMs.
18. Chang L, et al. CMOS circuit performance enhance- Trans IEICE 2003;J86-C(3):301–306. (in Japanese)
ment by surface orientation optimization. IEEE Trans 24. Jovanovic V, et al. Solid-State Electron 2010;54:870–
Electron Dev 2004;51(10):1621–1627. 876.
19. Sakurai T, Newton RA. Alpha-power law MOSFET 25. Jovanovic V, et al. Proceedings of Device Research
model and its application to CMOS inverter and other Conference 2009:261–262. (in Japanese)
Yu Hiroshima (nonmember) graduated from Shonan Institute of Technology (Grad. School of Engineering, Dept. of
Information Science) in 2010. Student research dealt with LSI design methods using double-gate type transistors and stacked
transistors (3D transistors). Now employed by Oi Electric Co., Ltd.
Takahiro Kodama (nonmember) graduated from Shonan Institute of Technology (Grad. School of Engineering, Dept. of
Information Science) in 2012. Student research dealt with LSI design methods using SGT, FinFET, and other 3D transistors.
Now employed by Japan Process Development Co., Ltd.
Shigeyoshi Watanabe (member) received a bachelor’s degree in measurement engineering from Keio University in 1977,
completed the M.E. program in applied physics at Tokyo Institute of Technology in 1979 and joined the Toshiba Center for
Semiconductor Research and Development. Research interests: nonvolatile memory and DRAM circuit design, devices and
circuits using micro CMOS, BiCMOS, SOI, 3D transistors, high-speed low-power circuit architecture for system LSIs, and so
on. Since 2005 professor at Shonan Institute of Technology (Dept. of Information Science) (D. Eng.).
62
Copyright of Electronics & Communications in Japan is the property of John Wiley & Sons,
Inc. and its content may not be copied or emailed to multiple sites or posted to a listserv
without the copyright holder's express written permission. However, users may print,
download, or email articles for individual use.