IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 61, NO. 8, AUGUST 2014 2711 Epitaxially Defined FinFET: Variability Resistant and High-Performance Technology Sushant Mittal, Shashank Gupta, Aneesh Nainani, Member, IEEE, Mathew C. Abraham, Member, IEEE, Klaus Schuegraf, Member, IEEE, Saurabh Lodha, Member, IEEE, and Udayan Ganguly, Member, IEEE Abstract— FinFET technology is prone to suffer from line edge roughness (LER)-based VT variation with scaling. It also lacks a simple implementation of multiple VT technology needed for power management. To address these challenges, in this paper we present an epitaxially defined FinFET (EDFinFET) as an alternate to FinFET architecture for nodes 15 nm and beyond. We show by statistical simulations that EDFinFET reduces overall VT variability with an 80% reduction in LER-based variability in comparison with FinFETs. We present dynamic threshold MOS (DTMOS) configuration of EDFinFET using the available body terminal to individual transistors. The DTMOS configuration reduces LER-based variability by 90% and overall variability by 59%. It also has excellent subthreshold slope (SS) and gives 43% higher ION compared with FinFETs. Meanwhile, EDFinFET shows poorer SS and lower ION than FinFET due to single gate control. However, it is capable of multiple VT , which leads to circuit level power optimization. Index Terms— DTMOS configuration of EDFinFET (DTEDFinFET), epitaxially defined FinFET (EDFinFET), FinFET, line edge roughness (LER), VT variability. I. I NTRODUCTION D EVICE variability is one of the major roadblocks for CMOS technology scaling [1]–[4]. The various sources of variability include random dopant fluctuation (RDF), gate edge roughness (GER), line edge roughness (LER), line width roughness (LWR), metal gate granularity and random telegraph noise [5]. The RDF is one of the major variability sources for planar devices [3], [6]. FinFETs with undoped fins have reduced the impact of RDF considerably [2]. However, in FinFETs the requirement to have a thin fin width (Wfin ∼ L G /3 [7]) for an adequate electrostatic control, increases the impact of LER on the device variability and makes it one of the most critical variability component [5]. The LER leads to roughness in fin width, (i.e., LWR), which results in increased quantum confinement (QC) effects in narrow sections of the Manuscript received March 7, 2013; revised April 30, 2014; accepted June 4, 2014. Date of publication June 24, 2014; date of current version July 21, 2014. This work was supported in part by the Department of Science and Technology, India, and in part by Applied Materials Inc., Santa Clara, CA, USA. The review of this paper was arranged by Editor J. C. S. Woo. S. Mittal, S. Lodha, and U. Ganguly are with the Department of Electrical Engineering, IIT Bombay, Mumbai 400 076, India (e-mail: smittal@iitb.ac.in; saurabh.lodha@gmail.com; udayan@ee.iitb.ac.in). S. Gupta, A. Nainani, M. C. Abraham, and K. Schuegraf are with Applied Materials Inc., Santa Clara, CA 94085 USA (e-mail: shashankgupta. iitb@gmail.com; aneesh_nainani@amat.com; mathew_abraham@amat.com; klaus_schuegraf@amat.com). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TED.2014.2329993 Fig. 1. VT shift for different gate lengths (nominal Wfin = L G /3) as Wfin is reduced from nominal. QC increases VT largely as Wfin gets reduced due to LER consistent with [8] at L G = 20 nm. fin. When QC is increased, VT variation increases. This effect is shown in Fig. 1 where the sensitivity of the threshold voltage of FinFETs on fin width from 18- to 9-nm node with different magnitudes of variation in fin width (representing the impact of LWR) is shown. The LWR magnitude implies 3σ variation of fin width, i.e., for nominal fin width of 5 nm, with LWR magnitude of 3 nm; effective fin width would be in between 2–8 nm (5 ± 3 nm). Larger VT variation would be for the fin width less than nominal fin width due to enhanced QC, which is shown in Fig. 1. As stated above, with gate length reduction, nominal fin width would also reduce for an adequate electrostatic control [7]. VT increases sharply as nominal fin width is decreased, which is consistent with literature [8]. In addition, line edge roughness reduction is a major challenge and has not scaled effectively with advanced nodes [9]. The FinFET technology lacks easy implementation of multiple threshold voltage (VT ) capability, which is a key enabler for power versus performance optimization by circuit designers [10]. In planar technology, this is achieved by various patterned implant steps, but such dopant profile engineering remains unavailable to FinFET technology due to its undoped fins. Processes including gate work function engineering, gate-drain/source overlap engineering [10], and so on have been proposed in simulations to achieve this multiple threshold voltage capability in FinFETs. These steps, 0018-9383 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information. 2712 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 61, NO. 8, AUGUST 2014 however, makes the fabrication process complex and therefore are undesirable. To address the variability concern and multiple VT issues for the FinFET technology, we propose an alternative to the conventional FinFET structure named the epitaxially defined FinFET (EDFinFET) [11]. In this paper, we demonstrate in detail the variability resistance and performance of the EDFinFET benchmarked against FinFETs. An easy multiple VT capability with just change of body bias was demonstrated earlier in [11]. The rest of this paper is organized as follows. Calibration of the Sentaurus simulation deck [12] and simulation methodology of different variability sources are described in Section II. In Section III, we introduce EDFinFET device architecture and propose a fabrication flow, followed by the optimization of the device structure. In Section IV, device performance comparison is done. The EDFinFET is statistically compared against FinFET for various sources of VT variability in Section V. Scalability is demonstrated in Section VI followed by conclusions in Section VII. Fig. 2. Calibration of simulation deck with experimental data for 25-nm channel length FinFET [14]. TABLE I PARAMETERS U SED FOR F IN FET S IMULATION II. C ALIBRATION OF S IMULATION D ECK AND D ESCRIPTION OF A PPROACH In TCAD Sentaurus test bench, following effects have been included. First, to capture current transport correctly, Masetti mobility model was used for impurity scattering [15] and Lombardi model was used for mobility degradation at interface [16]. To capture high-field saturation, extended Canali mobility model was used [17]. Above mentioned mobility models were combined using Matthiessen’s rule. Second, density gradient method was used to include QC effects accurately and efficiently in drift-diffusion (DD) simulations for nominal device and geometric variability studies [8]. Such quantum DD simulation underestimates ION at small channel lengths due to its inability to capture velocity overshoot effects [18]. Velocity overshoot effects were considered by quantum-corrected Monte Carlo (MC) simulations for the nominal device, (i.e., without geometrical variability effects like LER or GER). Quantum correction for classical MC simulation was introduced by appropriately changing Effective oxide thickness (EOT) and gate work function in DD classical simulation so that it matches DD with quantum effects, as explained in [12]. As MC simulation is computationally intensive, an approximation by increased saturation velocity was used as explained in [18] for large sample size geometric variability study. Third, to capture generation–recombination correctly, SRH recombination, trap assisted tunneling (TAT), and band to band tunneling (BTBT) was included. Scharfetter relation was used to consider doping dependence of recombination lifetime of SRH and Hurkx model was used for TAT [19]. Kane’s dynamic nonlocal model was used to account for BTBT leakage currents [20]. Finally, Oldslotboom model was used for bandgap narrowing [21], [22]. A 2-D model has been used in simulations consistent with FinFET models in [23]. The FinFET has two channels corresponding to the double gate architecture. Hence channel width is twice the fin height. As conventional, current per unit channel width (in mA/μm) for a fin is presented. Using the above mentioned models, TCAD simulation bench was calibrated to the experimental data at L G = 25 nm [14] (Fig. 2). Upon excellent matching to experiment as shown in Fig. 2, this wellcalibrated 2-D simulation deck was used to simulate variability and performance for FinFET and EDFinFET at 15-nm node. For 15-nm node FinFET, a fin width of 5 nm was used (∼L G /3 [7]) with an EOT of 0.67 nm, L SPACER of 5 nm, and VDD of 0.78 V [13]. A source/drain doping of 4 × 1020 was used along with 1 nm/decade drop to channel doping of 1 × 1017 . The parameters are summarized in Table I. To simulate LER- and GER-based variability, method presented in [5] has been adopted. Random lines with an exponential autocorrelation function (ACF) [9] with correlation length = 30 nm and 3 amplitude of 2 nm were generated [5]. is the standard deviation of the generated random line. The exponential ACF is a property of the solution of Langevin equation, which was used to generate random lines [24]. However, exponential ACF has high-frequency components and the same were removed by passing it through a low-pass filter (LPF) in MATLAB. The resultant ACF resembled Gaussian ACF as shown in [9]. Both, generated line and ACF are shown in Fig. 3(a) and (b), respectively, before and after LPF. A sample of FinFET structure generated using this methodology is shown in inset of Fig. 3(b). A sufficient ensemble size of 200 [2] was used for both LER- and GER-based variability evaluation of FinFET and EDFinFET devices using above described methodology. MITTAL et al.: ED FinFET: VARIABILITY RESISTANT AND HIGH-PERFORMANCE TECHNOLOGY 2713 Fig. 3. (a) Random line generated using Langevin equation and after passing through LPF. (b) Autocorrelation of random lines obtained for two cases. Correlation lengths for both are also shown. Inset: a sample FinFET structure with LER generated using this methodology. The FinFETs remain immune to RDF because of undoped fins. However, as we will see, EDFinFET requires doped fin. So estimation of RDF for EDFinFET is essential for which we used impedance field method (IFM) [25]. In this, the standard deviation in gate voltage (σ VG ) due to RDF to maintain nominal current is plotted versus nominal VG . σ VG at VT gives σ VT due to RDF. III. E PITAXIALLY D EFINED F IN FET A. Device Architecture and Fabrication Flow The FinFETs suffer from LER-based variability because fin is fully depleted or subjected to bulk inversion depending on bias conditions; and thus entire fin width contributes both in electrostatics and transport [e.g., enhanced quantization in narrow regions [Fig. 4(a.i)]]. This sensitivity to the fin width is the basis for the VT variability. To realize a FinFET which is insensitive to fin width, we propose a highly doped fin that is patterned first assuming a typical LER, as shown in Fig. 4(a). A low-doped channel is grown by conformal Si epitaxy on the highly doped fin with abrupt doping (∼1 nm/decade [26], [27]) [thus named epitaxially defined (ED) FinFET [11]]. Depletion width is defined by lightly doped epitaxy since underlying high-doped fin cannot be depleted. As lightly doped epitaxy remains uniform despite LER on the layer beneath [Fig. 4(a.ii)], depletion width remains constant. Therefore, inspite of having LER in the starting fin, depletion width is unaffected in EDFinFET. Thus, EDFinFET device architecture is expected to be largely free from LER-induced VT variations. Similar planar structure with retrograde doping [28], [29] has been demonstrated for reducing RDF-based variability. In this paper, we present a double gate adaptation to demonstrate LER-induced VT Fig. 4. (a) FinFET (a.i) Wfin = 5-nm subjected to LER. In FinFET, Wfin dependent QC is affected by LER-based Wfin variation. (a.ii) Equivalent EDFinFET with same extent of LER. In EDFinFET, QC is defined by the uniform low-doped epithickness, which is insensitive to LER; any variation due to LER is absorbed in heavily doped thick body. (b) Proposed process flow for EDFinFET. (c) 2-D cut of EDFinFET used in simulations. variations immunity, which is a critical issue for conventional FinFETs. The proposed fabrication flow of a p-channel EDFinFET is shown in Fig. 4(b). To fabricate EDFinFET, we can start with heavy n-doped fin (doping density ∼1020 cm−3 ) with fin width greater than 2–3 times the fin width of conventional FinFET (≥10 nm) on a p-well. A low-doped channel is grown by conformal Si epitaxy on the highly doped fin with abrupt doping (∼1 nm/decade [26], [27]). The gate is patterned next followed by spacer deposition. This step involves patterning dummy gates as well. Next, epitaxial layer between gate and dummy gates is etched out isotropically and partially in a self-aligned manner so that high-doped source and drain can be grown. This step is important since recessed source/drain is essential to reduce series resistance. Partial lightly doped epilayer is required to reduce BTBT leakage between highly doped body and drain as will be discussed later in this section. Excellent controls for partial isotropic etching can be achieved using plasma oxidation and Siconi™ etch process, which have digital etch control (<1 nm/cycle) [30], [31]. Following this, source/drain is grown by selective epitaxy. Self-aligned contacts for all the terminals of transistor are fabricated. A major simplification in the process flow is that in FinFET Wfin requires sublithographic patterning. However, in comparison, in EDFinFET fin is thicker and therefore lithography and 2714 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 61, NO. 8, AUGUST 2014 Fig. 5. Leakage current through body-drain reverse bias junction at V D = −0.78 V for different lightly doped epithickness between body and the drain as illustrated in the inset diagram of EDFinFET (T 2). T 2 higher than 2.7 nm can reduce BTBT leakage below 100 nA/μm. Fig. 6. ION and SS of EDFinFET for different Si epi thickness under gate (T 1) at T 2 = 3.5 nm. The SS degrades with increase in T 1 due to strong short channel effects. ION is a function of SS and source/drain series resistance. Source/drain series resistance decreases as T 1 increases since S/D junction depth (T 1–T 2) increases. Thus, there is an optimized ION point. Inset: the difference in biasing of (i) EDFinFET and (ii) DTEDFinFET. patterning requirements are relaxed. In addition, all proposed steps are same as that of FinFET except epitaxial deposition of channel layer and recessed growth of epitaxial source/drain. Epitaxial growth of channel has been demonstrated in [28] for a planar device and in [32] for a 3-D device. In addition, epitaxial source and drain is a standard step for DRAM [33] and FinFETs [7]. To use body-effect, individual transistor body-biasing can be enabled by providing a contact to the heavily doped fin (body) as well in the last fabrication step. Individual body contact is a unique capability of this architecture. In comparison, FinFET technology does not offer body effect advantage due to very low-body coefficients [34], while in planar technology body biasing is performed through well-contacts for multiple devices sharing the well. Therefore, individual transistor body biasing is not feasible for planar devices as well. Fig. 4(c) shows the final 2-D structure used for performance and variability evaluation. B. Performance Optimization of EDFinFET The lightly doped Si epi region in EDFinFET can be divided into two parts: 1) beneath the gate (marked as T 1 in inset of Fig. 5) and 2) beneath source and drain (marked as T 2 in inset of Fig. 5). T 1 has a strong effect on the device performance since it would determine control of gate on channel, while T 2 controls BTBT leakage between high-doped body and source/drain. Impact of these parameters is presented. To estimate BTBT current between n+ body and p+ source/drain, which depends on T 2, we have used Kane nonlocal model [20] after testing its calibration with experimental data in [19]. With this model, the leakage current between reverse biased body–drain junctions for different T 2 is shown in Fig. 5. T 2 needs to be larger than 2.7 nm to reduce leakage current below 100 nA/μm. To be conservative, we choose it to be 3.5 nm in our simulations. Fig. 6 shows effect of T 1 on ION (at IOFF = 100 nA/μm) and SS for EDFinFET at T 2 = 3.5 nm. Poor subthreshold slope and ION of EDFinFET is due to single side gate control, which can be improved significantly using EDFinFET in DTMOS configuration [35] where in gate bias is simultaneously applied to the body terminal [configuration we refer to as DTMOS configuration of EDFinFET (DTEDFinFET)]. The DTEDFinFETs subthreshold slope of 70 mV/decade is comparable with FinFET. Consequently, DTEDFinFET has higher ION than EDFinFET. In both configurations, when T 1 is large source/drain fields easily penetrate through the lightly doped epilayer, (i.e., strong short-channel effects) leading to poor subthreshold slope [36]. When T 1 is small, the S/D junction depth which is T 1–T 2, is small. This increases S/D series resistance and degrades ION for lower T 1. These two effects produce an optimum range of T 1 for maximum ION . We observe that the ION for DTEDFinFET is maximum for T 1 around 6 nm and then degrades for smaller or larger T 1. A similar trend is observed for EDFinFET with optimum T 1 of 11 nm. As DTEDFinFET shows higher ION than EDFinFET, we choose T 1 = 6 nm for the rest of the simulation study for performance and variability comparison with FinFET. Thus, optimized T 1 of 6 nm, T 2 of 3.5 nm for the 1017 cm−3 n-doped epilayer and nominal T 3 (marked in inset of Fig. 5) of 10 nm for 1020 cm−3 n-doped central fin was used in simulations throughout at 15-nm node. Other parameters are same as FinFET as shown in Table I. IV. P ERFORMANCE C OMPARISON I D –VG characteristics of EDFinFET are compared against FinFET in Fig. 7. Linear scale curve is obtained by quantumcorrected Monte Carlo simulations without stress dependent mobility enhancement, wherein EOT and gate work function is adjusted so that classical simulations mimic quantum simulations. Quantum simulations used density gradient MITTAL et al.: ED FinFET: VARIABILITY RESISTANT AND HIGH-PERFORMANCE TECHNOLOGY 2715 Fig. 7. Performance comparison of EDFinFET and FinFET–EDFinFET shows slightly lower SS due to single gate control leading to poor ION . Quantum-corrected Monte Carlo (MC) simulations results are shown on right y-axis in linear scale. DTEDFinFET configuration restores the poor SS and also improves the ION by 43%. Inset: DTEDFinFETs body current is negligible compared with drain current during operation. approximation as is mentioned above. This is presented to consider velocity overshoot effects, while comparing ION for the three devices. The EDFinFET because of single side gate control has slightly poor SS, leading to a poor ION for same IOFF . However, for complete comparison of ION , stress effects are required to be included, and EDFinFET, apart from regular source–drain uniaxial stress, has an additional feasibility of applying biaxial stress through high-doped body beneath. This additional stress can produce 10× mobility enhancement [37] and therefore can potentially reduce the difference between ION of EDFinFET and FinFET [38]. On the other hand, DTEDFinFET configuration leads to large improvement in SS and ION , as shown in Fig. 7. This is attributed to the fact that DTEDFinFET is essentially a dual gate device as body bias assists the gate control compared with EDFinFET. The DTEDFinFETs SS is comparable with that of FinFET and ION is 43% higher. Though the body–source junction becomes forward-biased with increase in body bias, the body leakage is well below the drain current, as shown in the inset of Fig. 7, in consistency with DTMOS literature [35]. V. VT VARIABILITY C OMPARISON Fig. 8 compares statistically the impact of LER variation on FinFET, EDFinFET, and DTEDFinFET. Fig. 8(a) shows for the three devices raw I D –VG plots of 200 samples along with nominal device I D –VG . Nominal I D –VG has same IOFF of 100 nA/μm. Constant-current method was used to calculate VT at 10−2 mA/μm and is marked in Fig. 8(a). Variation in fin width, leading to QC, causes FinFET I D –VG to vary significantly. However, both EDFinFET and DTEDFinFET are immune to such variation due to two additive effects. First, the depletion width of channel is uniform as discussed earlier in the text. Second, single dielectric interface forms weaker (triangular) well and reduce QC compared with FinFETs with double dielectric interface forming a square well. Fig. 8(b) Fig. 8. (a) Variation in I D –VG of FinFET, EDFinFET, and DTEDFinFET due to LER. FinFET shows a lot of variation. (b) Q–Q plot of VT distribution due to LER. As shown, FinFET get affected by QC leading to deviation from normal distribution. EDFinFET removes QC and hence vary very less. DTEDFinFET behaves ideally with almost no LER-based variability due to no QC and symmetric gate control. shows Q–Q plot of VT distribution obtained from I D –VG , as shown in Fig 8(a). For comparison, nominal VT is matched at 0 V for three devices. As shown, QC leads FinFET curve to deviate from normal distribution, in consistency with [5]. Both EDFinFET and DTEDFinFET show normal distribution and less variation in comparison with FinFETs, as explained above. The DTEDFinFET shows even lesser variation. This is because shift in VT observed for EDFinFET was due to asymmetric gate control compared with the body terminal (inset of Fig. 6), leading to either enhancement of gate control [convex or thicker fins—Fig. 8(b-i)] or loss of gate control [concave or thinner fins—Fig. 8(b-ii)]. Gate terminal shorted to body removes this asymmetric nature of gate control as a concave gate comes with a convex highly doped body, which shares the gate bias and vice versa. Hence effective gate control does not change with LER, leading to much improved variability immunity and making it almost completely 2716 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 61, NO. 8, AUGUST 2014 Fig. 10. Comparison of overall σ VT for FinFET, EDFinFET, and DTEDFinFET. EDFinFET gives 35% reduction in overall variability. DTEDFinFET reduces LER by 90% and overall variability by 59%. Fig. 9. (a) Q–Q plot of VT distribution for three devices due to GER. EDFinFET is poorer than FinFETs due to single gate control. DTEDFinFET configuration improves it and makes it comparable with FinFET. Inset: cartoon of EDFinFET with GER (a.i) with increased L G (L G1 and L G2 ) and (a.ii) with reduced L G (L G3 and L G4 ). (b) Comparison of σ VT due to RDF by IFM [25]: the standard deviation in gate voltage (σ VG ) for a constant drain current at nominal VG due to RDF is plotted as σ VG versus VG . σ VG is extracted at VT to obtain σ VT . VT is extracted by constant current method at I D = 10−2 mA/μm and is marked in (b). resistant to LER-based variability with a low σ VT of 3.4 mV. This configuration is similar to a FinFET with LWR equal to zero. In Fig. 9(a), Q–Q plot of VT distribution obtained due to GER for three devices is compared, while inset shows cartoon of simulated EDFinFET with GER: 1) with increased L G (marked as L G1 and L G2 ) and 2) with reduced L G (marked as L G3 and L G4 ). Here, also for comparison, nominal VT is matched at 0 V for three devices. FinFET, with good electrostatic control of channel, varies less with a σ VT of 7.4 mV only. Due to single gate control, EDFinFET performs poor than FinFETs and has 1.8× higher σ VT . However, DTEDFinFET configuration owing to better channel control shows improvement over EDFinFET and has σ VT comparable to FinFETs. Fig. 9(b) compares the RDF VT fluctuation for the three devices. As shown, despite having high doping in fins, σ VT of EDFinFET is comparable with that of FinFETs. This is attributed to the fact that since the depletion width is limited to lightly doped epitaxial channel and high-doped fins remains undepleted, the dopant fluctuations in high-doped fins does not contribute to VT variability and remains screened out. It is consistent with planar retrograde doping profile devices [28]. σ VT by three independent random sources is calculated above. Variance of three independent random sources adds to give overall variance. Overall σ VT is square root of overall variance. Thus, obtained individual and overall comparison of σ VT due to these variability sources is shown in Fig. 10. 80% reduction in variability due to LER and 35% in overall variability are achieved by EDFinFET. However, DTEDFinFET configuration is capable of achieving a significant reduction of 90% in LER variability and 59% in overall variability. Thus, EDFinFET and DTEDFinFET configurations are much beneficial to reduce variability. Fig. 11 shows the VT variability and performance tradeoff of FinFETs versus EDFinFET technology. FinFET variability maybe improved at the cost of performance by increasing the nominal fin width. In comparison for similar variability, DTEDFinFET has 43% higher ION with 59% less VT variability. EDFinFET has 35% low-VT variability, but ION also is less. This loss can possibly be compensated by biaxial stress benefit exclusively available in EDFinFET [37], [38]. In addition, Fig. 11 also shows sudden increase in VT variability of FinFET due to further reduction of Wfin, posing question on feasibility of Wfin = L G /3 thumb rule of acceptable short-channel control of channel at future technology nodes. VI. S CALABILITY At 10-nm node, with the rule of Wfin ∼ L G /3 which is required to maintain good electrostatics, due to a further increase in LER variability, FinFETs become all the more variable as shown in Fig. 11. So by fixing Wfin of FinFETs MITTAL et al.: ED FinFET: VARIABILITY RESISTANT AND HIGH-PERFORMANCE TECHNOLOGY 2717 ACKNOWLEDGMENT The authors would like to thank Prof. A. Tulapurkar, Prof. S. Mahapatra, Prof. S. Ganguly, and Dr. A. B. Sachid for discussions. R EFERENCES Fig. 11. Comparison of σ VT with ION for FinFET, EDFinFET, and DTEDFinFET. DTEDFinFET has 59% lower σ VT with 43% higher ION . In addition, sudden further increase in variability of FinFET for lesser Wfin of 3.5 nm is also shown. TABLE II S CALABILITY OF EDF IN FET C OMPARED W ITH F IN FET to be equal to 5 nm for 10-nm node as well, scalability data is shown in Table II. Epi thickness of EDFinFET’s channel (T 1 in inset of Fig. 5) has been optimized as discussed in Section III-B for 10-nm node also. EDFinFET still has its inherently poorer ION , SS, and DIBL. The DTEDFinFET, on the other hand, gives greater ION boost, good SS, and comparable DIBL at 10-nm node as well. VII. C ONCLUSION In this paper, we propose EDFinFET device architecture as an alternate to conventional FinFET architecture. Depletion width of EDFinFET, being defined by epitaxy, remains constant inspite of LER on the patterned fin. This leads to 80% reduction in LER-based variability and an overall reduction of 35% over FinFETs. The DTMOS configuration of EDFinFET (DTEDFinFET) removes asymmetric nature of the device and shows 90% reduction in LER VT variability, with an overall reduction of 59%. From a performance perspective, single gate control leads to poor SS, DIBL, and ION for EDFinFET. Due to increased gate control, DTEDFinFET restores SS and shows 43% ION boost over FinFETs along with comparable DIBL. We also showed that the EDFinFET logic is scalable and DTEDFinFET promises even better performance at 10-nm node compared with FinFETs. [1] X. Wang, A. R. Brown, N. Idris, S. Markov, G. Roy, and A. Asenov, “Statistical threshold-voltage variability in scaled decananometer bulk HKMG MOSFETs: A full-scale 3-D simulation scaling study,” IEEE Trans. Electron Devices, vol. 58, no. 8, pp. 2293–2301, Aug. 2011. [2] E. Baravelli, M. Jurczak, N. Speciale, K. De Meyer, and A. Dixit, “Impact of LER and random dopant fluctuations on FinFET matching performance,” IEEE Trans. Nanotechnol., vol. 7, no. 3, pp. 291–298, May 2008. [3] A. Asenov, “Random dopant induced threshold voltage lowering and fluctuations in sub 50 nm MOSFETs: A statistical 3D ‘atomistic’ simulation study,” Nanotechnology, vol. 10, no. 2, pp. 153–158, 1999. [4] Y. Li and C.-H. Hwang, “Discrete-dopant-induced characteristic fluctuations in 16 nm multiple-gate silicon-on-insulator devices,” J. Appl. Phys., vol. 102, no. 8, pp. 084509-1–084509-6, Oct. 2007. [5] X. Wang, A. R. Brown, B. Cheng, and A. Asenov, “Statistical variability and reliability in nanoscale FinFETs,” in Proc. IEEE Int. Electron Devices Meeting (IEDM), Dec. 2011, pp. 5.4.1–5.4.4. [6] Y. Li, S.-M. Yu, J.-R. Hwang, and F.-L. Yang, “Discrete dopant fluctuations in 20-nm/15-nm-gate planar CMOS,” IEEE Trans. Electron Devices, vol. 55, no. 6, pp. 1449–1455, Jun. 2008. [7] J. Kedzierski et al., “Extension and source/drain design for highperformance FinFET devices,” IEEE Trans. Electron Devices, vol. 50, no. 4, pp. 952–958, Apr. 2003. [8] S. Xiong and J. Bokor, “Sensitivity of double-gate and FinFETDevices to process variations,” IEEE Trans. Electron Devices, vol. 50, no. 11, pp. 2255–2261, Nov. 2003. [9] A. Asenov, S. Kaya, and A. R. Brown, “Intrinsic parameter fluctuations in decananometer MOSFETs introduced by gate line edge roughness,” IEEE Trans. Electron Devices, vol. 50, no. 5, pp. 1254–1260, May 2003. [10] S. A. Tawfik and V. Kursun, “Multi-threshold voltage FinFET sequential circuits,” IEEE Trans. Very Large Scale Integr. (VLSI) Syst., vol. 19, no. 1, pp. 151–156, Jan. 2011. [11] S. Mittal et al., “Epitaxialy defined (ED) FinFET: To reduce VT variability and enable multiple VT,” in Proc. 70th Annu. Device Res. Conf. (DRC), Jun. 2012, pp. 127–128. [12] Sentaurus TCAD Design Suite [Online]. Available: http://www.synopsys.com [13] (2009). ITRS Roadmap [Online]. Available: http://www.itrs.net/ Links/2009ITRS/Home2009.htm [14] C.-Y. Chang et al., “A 25-nm gate-length FinFET transistor module for 32 nm node,” in Proc. IEEE Int. Electron Devices Meeting (IEDM), Dec. 2009, pp. 1–4. [15] G. Masetti, M. Severi, and S. Solmi, “Modeling of carrier mobility against carrier concentration in arsenic-, phosphorus-, and boron-doped silicon,” IEEE Trans. Electron Devices, vol. 30, no. 7, pp. 764–769, Jul. 1983. [16] C. Lombardi, S. Manzini, A. Saporito, and M. Vanzi, “A physically based mobility model for numerical simulation of nonplanar devices,” IEEE Trans. Comput.-Aided Design Integr. Circuits Syst., vol. 7, no. 11, pp. 1164–1171, Nov. 1988. [17] C. Canali, G. Majni, R. Minder, and G. Ottaviani, “Electron and hole drift velocity measurements in silicon and their empirical relation to electric field and temperature,” IEEE Trans. Electron Devices, vol. 22, no. 11, pp. 1045–1047, Nov. 1975. [18] R. Granzner et al., “Simulation of nanoscale MOSFETs using modified drift-diffusion and hydrodynamic models and comparison with Monte Carlo results,” Microelectron. Eng., vol. 83, no. 2, pp. 241–246, Feb. 2006. [19] G. A. M. Hurkx, D. B. M. Klaassen, and M. P. G. Knuvers, “A new recombination model for device simulation including tunneling,” IEEE Trans. Electron Devices, vol. 39, no. 2, pp. 331–338, Feb. 1992. [20] E. O. Kane, “Theory of tunneling,” J. Appl. Phys., vol. 32, no. 1, pp. 83–91, 1961. [21] J. W. Slotboom and H. C. de Graaff, “Measurements of bandgap narrowing in Si bipolar transistors,” Solid-State Electron., vol. 19, no. 10, pp. 857–862, Oct. 1976. 2718 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 61, NO. 8, AUGUST 2014 [22] J. W. Slotboom and H.C. de Graaff, “Bandgap narrowing in silicon bipolar transistors,” IEEE Trans. Electron Devices, vol. 24, no. 8, pp. 1123–1125, Aug. 1977. [23] T. Krishnamohan, “Band-engineering of novel channel materials for high performance nanoscale MOSFETs,” in Proc. Int. Conf. Simul. Semicond. Process. Devices (SISPAD), Sep. 2008, pp. 97–100. [24] J. L. García-Palacios and F. J. Lázaro, “Langevin-dynamics study of the dynamical properties of small magnetic particles,” Phys. Rev. B, vol. 58, no. 22, pp. 14937–14958, Dec. 1998. [25] A. Wettstein, O. Penzin, E. Lyumkis, and W. Fichtner, “Random dopant fluctuation modelling with the impedance field method,” in Proc. Int. Conf. Simul. Semicond. Process. Devices, Sep. 2003, pp. 91–94. [26] H.-J. Gossmann, F. C. Unterwald, and H. S. Luftman, “Doping of Si thin films by low-temperature molecular beam epitaxy,” J. Appl. Phys., vol. 73, no. 12, pp. 8237–8241, 1993. [27] Y. Morita et al., “Synthetic electric field tunnel FETs: Drain current multiplication demonstrated by wrapped gate electrode around ultrathin epitaxial channel,” in Proc. Symp. VLSI Technology (VLSIT), Jun. 2013, pp. T236–T237. [28] K. Fujita et al., “Advanced channel engineering achieving aggressive reduction of VT variation for ultra-low-power applications,” in Proc. IEEE Int. Electron Devices Meeting (IEDM), Dec. 2011, pp. 32.3.1–32.3.4. [29] X. Sun et al., “Tri-gate bulk MOSFET design for CMOS scaling to the end of the roadmap,” IEEE Electron Device Lett., vol. 29, no. 5, pp. 491–493, May 2008. [30] U. Ganguly et al., “Scalability enhancement of FG NAND by FG shape modification,” in Proc. IEEE Int. Memory Workshop (IMW), May 2010, pp. 1–4. [31] S. Kubicek et al., “Bulk FinFET fabrication with new approaches for oxide topography control using dry removal techniques,” Solid-State Electron., vol. 71, pp. 106–112, May 2012. [32] D. Cohen-Elias et al., “Formation of sub-10 nm width InGaAs finFETs of 200 nm height by atomic layer epitaxy,” in Proc. 71st Annu. Device Res. Conf. (DRC), Notre Dame, IN, USA, Jun. 2013, pp. 1–2. [33] J.-S. Park et al., “Mobility enhancement of peripheral PMOSFET using e-SiGe source and drain in sub-50 nm DRAM,” in Proc. 4th IEEE Int. Memory Workshop (IMW), May 2012, pp. 1–4. [34] J.-W. Han, C.-H. Lee, D. Park, and Y.-K. Choi, “Body effects in tri-gate bulk FinFETs for DTMOS,” in Proc. IEEE Nanotechnol. Mater. Devices Conf. (NMDC), vol. 1. Oct. 2006, pp. 208–209. [35] F. Assaderaghi, D. Sinitsky, S. A. Parhe, J. Bahor, P. K. Ko, and C. Hu, “Dynamic threshold-voltage MOSFET (DTMOS) for ultra-low voltage VLSI,” IEEE Trans. Electron Devices, vol. 44, no. 3, pp. 414–422, Mar. 1997. [36] Y. Taur and T. H. Ning, Fundamentals of Modern VLSI Devices. Cambridge, U.K.: Cambridge Univ. Press, 1998. [37] M. L. Lee, E. A. Fitzgerald, M. T. Bulsara, M. T. Currie, and A. Lochtefeld, “Strained Si, SiGe, and Ge channels for high-mobility metal-oxide-semiconductor field-effect transistors,” J. Appl. Phys., vol. 97, no. 1, pp. 011101-1–011101-27, 2005. [38] S. Mittal et al., “Epi defined (ED) FinFET: An alternate device architecture for high mobility Ge channel integration in PMOSFET,” in Proc. IEEE 5th Int. Nanoelectron. Conf. (INEC), Singapore, Jan. 2013, pp. 367–370. Authors’ photographs and biographies not available at the time of publication.