The Impact of Layout-Dependent STI Stress and Effective Width on Low-Frequency Noise and High-Frequency Performance in Nanoscale nMOSFETs

全文

(1)

and High-Frequency Performance in

Nanoscale nMOSFETs

Kuo-Liang Yeh, Member, IEEE, and Jyh-Chyurn Guo, Senior Member, IEEE

Abstract—The impact of channel width scaling on

low-frequency noise (LFN) and high-low-frequency performance in multi-finger MOSFETs is reported in this paper. The compressive stress from shallow trench isolation (STI) cannot explain the lower LFN in extremely narrow devices. STI top corner rounding (TCR)-inducedΔW is identified as an important factor that is responsible for the increase in transconductance Gm and the reduction in LFN with width scaling to nanoscale regime. A semi-empirical model was derived to simulate the effective mo-bility eff) degradation from STI stress and the increase in

effective width(Weff) from ΔW due to STI TCR. The proposed

model can accurately predict width scaling effect onGm based on a tradeoff betweenµeff andWeff. The enhanced STI stress

may lead to an increase in interface traps density (Nit), but

the influence is relatively minor and can be compensated by the

Weff effect. Unfortunately, the extremely narrow devices suffer

fT degradation due to an increase in Cgg. The investigation of

impact from width scaling onµeff,Gm, and LFN, as well as the

tradeoff between LFN and high-frequency performance, provides an important layout guideline for analog and RF circuit design.

Index Terms—Effective mobility, effective width, low-frequency

noise (LFN), shallow trench isolation (STI) stress.

I. INTRODUCTION

W

ITH the aggressive scaling of CMOS technology into the nanoscale regime, the stress introduced from shal-low trench isolation (STI) process becomes significant and im-poses a dramatic impact on not only CMOS device performance in dc current and ac gate speed [1]–[6] but also high frequency and analog performance. Among previous work on this subject, most of research effort and publications have been focused on the longitudinal stress (σ//) from STI along the direction of the

channel length [1]–[3]. However, relatively fewer studies were done for looking into the influence from STI transverse stress ) along the gate width and transverse to the channel length [4]–[6]. In addition, most of the studies on σ are limited to dc characteristics, such as threshold voltage (VT) and channel

Manuscript received May 10, 2010; accepted August 9, 2010. Date of publication October 4, 2010; date of current version November 5, 2010. This work was supported in part by the National Science Council under Grant NSC 98-2221-E009-166-MY3. The review of this paper was arranged by Editor Z. Celik-Butler.

The authors are with the Institute of Electronics Engineering, National Chiao Tung University, Hsinchu 300, Taiwan (e-mail: jcguo@mail.nctu.edu.tw).

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.2010.2072959

current (IDS), and restricted to long-channel devices, assuming

negligible σ// in the channel region [4]. STI edge effect on

flicker noise was investigated, and negative impact was reported for NMOS with narrow widths [7], [8]. A minor layout modifi-cation, i.e., an edge-extended design, was implemented, trying to reduce the stress and traps introduced by STI and fix the problem of non-1/f characteristics in the measured flicker noise [9]. However, the abnormal deviation from 1/f characteristics as reported for narrow devices cannot be reproduced from our experiment, even with a much more aggressively scaled width. A ring-type device was proposed, trying to eliminate σ⊥

and identify the influence on flicker noise [10]. However, the study is limited to single-polygate MOSFETs, which are not suitable for analog and RF circuit design, and the impact on high-frequency performance is unknown [10]. Recent studies reported a turn around of IDSwith decreasing width of the

ac-tive area and proposed an increase in effecac-tive width, i.e., ΔW , from STI top corner rounding (TCR) as the potential factor trad-ing with mobility degradation from STI compressive stress [4], [5]. Again, most of the works were restricted to single-polygate MOSFETs for logic circuits, but the influence on multiple-polygate-finger (i.e., multifinger) MOSFETs required for RF and analog circuits, with special concern of high-frequency performance and flicker noise, remains an open question.

The aforementioned multifinger MOSFETs have been widely used to reduce the parasitic gate resistance (Rg) in RF

circuits for the purpose of improving analog and RF perfor-mance like higher maximum oscillation frequency (fMAX) and

lower RF noise [11]. Note that the multifinger transistor has been recognized as the standard layout for RF circuit design. In this paper, two new layouts derived from the standard one, i.e., narrow-OD and multi-OD, were implemented, trying to enhance the transverse stress (σ) from STI. The primary ob-jective is to investigate STI transverse stress effect on effective mobility (μeff), transconductance (Gm), cutoff frequency (fT),

fMAX, and, most importantly, the low-frequency noise (LFN)

for both RF and analog circuit design.

II. DEVICEFABRICATION ANDCHARACTERIZATION

In this paper, multifinger MOSFETs were fabricated in 90-nm low-leakage CMOS process with nitrided oxide of target physical thickness at 2.2 nm. The gate length drawn on the

(2)

(b) Narrow-OD device:WF× NF= 1 μm × 32 (W1N32) and 0.5 μm × 64 (W05N64). (c) Multi-OD devices:WOD× NOD= 2 μm × 1, 0.25 μm × 8,

and 0.125μm × 16.

layout is 90 nm, i.e., Ldrawn= 90 nm, and the total channel

width Wtot is fixed at 32 μm. In order to investigate the stress

and interface traps generated from the STI process, two kinds of new layouts derived from multifinger MOSFET, i.e., multi-OD and narrow-multi-OD, were designed and implemented. Note that OD means oxide diffusion, which is equivalent to active area (generally denoted as AA). Fig. 1(a)–(c) displays the device layouts for standard, narrow-OD, and multi-OD, respectively. The narrow-OD devices illustrated in Fig. 1(b) were designed with simultaneously varied NF and WF under a specified total

width, i.e., Wtot = WF × NF. For this group of structures, WF× NF = 2 μm × 16, 1 μm × 32, and 0.5 μm × 64

cor-responding to finger width Wtot= 32 μm were fabricated in

this work. The multi-OD devices shown in Fig. 1(c) represent multiple-OD fingers with simultaneously varied OD finger width (WOD) and OD finger number (NOD) under a specified

finger width, which is WF = WOD× NOD. For this category

of devices, the polygate finger number NF is fixed at 16, and

WOD× NOD= 2 μm × 1, 0.25 μm × 8, and 0.125 μm × 16,

corresponding to finger width WF = 2 μm. Note that the

polygate-edge/OD-edge distance along the direction of channel length is fixed at 0.5 μm for both narrow-OD and multi-OD devices.

S-parameters were measured by an Agilent network analyzer E8364B for high-frequency characterization and ac parameter extraction. Open de-embedding was performed to remove the parasitic capacitances from the pads and interconnection lines, and short de-embedding was done to eliminate the parasitic resistances and inductances originated from the metal inter-connection. Fig. 2 demonstrates the LFN measurement system, which consists of an Agilent dynamic signal analyzer (DSA 35670) and low-noise amplifier (LNA SR570). The aforemen-tioned system was employed to measure the power spectral density of drain current noise, i.e., SID, which is the so-called

LFN. In this paper, the LFN was measured from multifinger MOSFETs under various gate voltages (VGS) and fixed drain

voltage at VDS= 50 mV. Note that VGS is converted as gate

overdrive, i.e., VGT= VGS− VT to offset VT variation among

different devices. The LFN measurement generally covers a wide frequency range of 4 Hz–10 kHz. The aforementioned measurement for dc and LFN characterization has been carried out on 10–15 dies for a statistical analysis.

The charge-pumping (CP) current system was built up, as shown in Fig. 3(a), for interface trap density (Nit)

measure-ment. The CP method proposed by Elliot [12] with sweeping base and fixed amplitude was applied to extract Nit from the

Fig. 2. LFN measurement system setup consisting of an Agilent dynamic signal analyzer (DSA 35670), low noise amplifier (LNA SR570), and Agilent 4156B for dc power supply. The measurement is automatically controlled by Agilent ICCAP.

Fig. 3. (a) CP measurement system consisting of dc source Agilent 4156B, pulse generator 81110A, and switching matrix Keithley 707A. (b) CP pulse waveform, wheretrandtfare the rising and falling times, andVbaseandVh

are swept from accumulation(Vbase< VFB) to inversion (Vh> VT), under

fixed pulse amplitudeVa= Vh− Vbase.

maximum CP current (ICP,max). Fig. 3(b) illustrates the pulse

waveform employed for the Elliot method, in which the base level (Vbase) is swept from accumulation to inversion while

keeping the pulse amplitude (Va) constant. Note that ICP,max

can occur under the condition of Vbase< VFB and Vbase+ Va = Vh> VT, and the accuracy is justified with a clear plateau for ICP,maxand a precise linear dependence on amplitude (Va)

and frequency (not shown).

III. LOCALSTRAINENGINEERINGEFFECT ON

NMOS PERFORMANCE

The STI stress introduced in MOSFETs with three different layouts as previously mentioned (standard, narrow-OD, and multi-OD) are illustrated in Fig. 4 to assist an analysis and understanding of layout effect on STI stress and, then, the electrical characteristics. Note that STI stress is classified as longitudinal stress, which is denoted as σ// and in parallel

with the channel length, and transverse stress, i.e., σ, which is transverse to the channel length. In this paper, the longitudinal stress σ//is considered to be similar for all of the devices with

various layouts, due to fixed gate length and polygate-edge/OD-edge distance [13]. Regarding the stress favorable for mobility enhancement, it has a critical dependence on the device types and orientations. The experimental results in previous work in-dicate that tensile stress in the longitudinal direction (σ//) can

enhance electron mobility in NMOS, whereas it will degrade hole mobility in PMOS. As for the transverse stress σof our

(3)

Fig. 4. Schematic of STI stresses along the longitudinal and transverse directions, which are defined asσ//andσ⊥in MOSFETs with three different layouts. (a) Standard multifinger device. (b) Narrow-OD device. (c) Multi-OD devices.

major interest in this paper, the compressive stress from STI is expected to degrade the effective mobility μefffor both NMOS

and PMOS [6].

A. STI Transverse Stress Effect on DC Performance of Multi-OD and Narrow-OD NMOS

As previously mentioned, narrow-OD and multi-OD MOSFETs were designed to study OD width scaling effect on transverse stress σfrom STI in different layouts. Narrow-OD devices are shown in Fig. 4(b), wherein the Narrow-OD width (WOD= WF) is reduced to be extremely narrow and the

polygate finger number (NF) is simultaneously increased to

keep the total width (Wtot= WF × NF) unchanged.

Multi-OD devices illustrated in Fig. 4(c) features a matrix of multiple ploy fingers, as well as multiple OD fingers (NOD) with narrow

OD width (WOD). Note that WODand NODare simultaneously

varied to keep WOD× NOD the same as the finger width of

standard device, i.e., WOD× NOD= WF = 2 μm.

It has been known that the STI process generally leads to VT

lowering with channel width scaling, and it is the so-called in-verse narrow-width effect (INWE) [14]. As shown in Fig. 5, the

VT versus WODfor narrow-OD and multi-OD NMOS presents

an obvious INWE, following a universal curve for different layouts. Note that INWE is determined by collective effects from STI TCR (ΔW ), STI stress, corner field crowding, doping profile, etc. Considering VTvariations from the aforementioned

effects, VGT= VGS− VTis used for electrical characterization

and analysis.

Fig. 6(a) presents the transconductance (Gm) under varying

VGT’s, which are measured from narrow-OD NMOS with two

splits of WF× NF, such as 1 μm× 32 (W1N32) and 0.5 μm ×

64 (W05N64) and the standard one (W2N16) for compari-son. The result indicates that the smaller WOD(= WF) leads

to lower Gm and the maximum Gm (Gm,max) in W05N64

(WOD= WF = 0.5 μm) is degraded by about 8%, compared

with the standard one, i.e., W2N16 (WOD= WF = 2 μm), as

shown in Fig. 6(b). The monotonic degradation of Gm with

Fig. 5. LinearVTversusWODfor narrow-OD and multi-OD NMOSs under

the biases ofVDS= 50 mV and VBS= 0 V.

Fig. 6. (a) Transconductance Gm versus VGT. (b) Maximum

transcon-ductance Gm_max measured from narrow-OD NMOS W1N32 (WF =

1 μm, NF= 32) and W05N64 (WF = 0.5 μm, NF= 64), and standard multifinger device W2N16(WF = 2 μm, NF = 16). All of the devices have the same total finger widthWF× NF = 32 μm.

Fig. 7. (a) TransconductanceGmversusVGT. (b) Maximum

transconduc-tanceGm_maxmeasured from multi-OD NMOS OD8(NOD= 8, WOD=

0.25μm) and OD16 (NOD= 16, WOD= 0.125 μm), and standard

multifin-ger device OD1(NOD= 1, WOD= 2 μm). All of the devices have the same

polygate finger number(NF = 16) and total OD width along each gate finger ((NOD× WOD= 2 μm).

WOD scaling in narrow-OD devices suggests that the increase

in STI compressive σ is the dominant factor responsible for

μeff degradation and the resulting Gm degradation. As for

multi-OD NMOS shown in Fig. 7, the Gm,maxof OD8 (NOD=

8, WOD = 0.25 μm) is degraded by about 20%, compared with

the standard one (OD1), but the continuous scaling of WOD

to 0.125 μm in OD16 (NOD= 16, WOD= 0.125 μm) leads

to an increase in Gm, compared with OD8, and the Gm,max

degradation, compared with the standard one, is shrunk to 11%. The result looks very interesting and cannot be explained by STI compressive stress alone. Note that the statistical variations of VT and Gm remain less than 2% from measurement on 10–15 dies, and it ensures the experimental results a high confidence level.

(4)

Δμ

μ0 = −(k⊥

σ+ k//σ//) (1)

where

μ0 mobility free from STI-stress-induced degradation;

Δμ mobility variation due to STI stress;

σ transverse stress perpendicular to the direction of the channel length;

σ// longitudinal stress along with the direction of the channel

length;

k first order of coefficient expressing mobility variation proportional to σ;

k// first order of coefficient expressing mobility variation

proportional to σ//. Then Δμ μ0 =−k⊥σ⊥ (2)

where σis a function of gate width expressed by

σ= −k · log(WOD) + k0 (3)

and can be written as another expression given by

σ= k · log W ref WOD  (4) where WOD× NOD= WF. (5)

In the preceding expressions, both longitudinal stress σ// and

transverse stress σ⊥ are considered in the original model. For multifinger MOSFETs, σ// limits its effect to the polygate at

the two ends. Therefore, σ//  σ⊥for all of the polygates other

than those at the two ends and it leads to the approximation of (1) to (2).

According to (3) or (4), it is proposed that Gmshould

de-crease with the reduction in WODsince the mobility decreases

as the transverse compressive stress σ increases with WOD

scaling.

However, in this work, we found that OD width scaling from WOD= 0.25 μm in OD8 to WOD= 0.125 μm in OD16

led to an increase in Gm,max, as shown in Fig. 7, rather than

degradation predicted by the stress model in (1)–(5). This observation suggests that the transverse compressive stress σ⊥

from STI, which is maximized in OD16 due to the minimum

WOD, cannot fully explain the largest Gm,max degradation in

OD8, instead of OD16. The experimental results suggest that the variation of Gm with WOD scaling is determined by not

only STI stress effect on mobility (μeff) but also STI TCR

Leff Gm=∂V∂IDS GS = WeffCoxμeff VDS Leff . (7) Let β = Cox VDS Leff (8) μeff(WOD) = μ0+ Δμ(WOD) = μ0  1− k · log W ref WOD  (9) Weff = (WOD+ ΔW ) × NOD× NF. (10) Then Gm= βμ0  1− k · log W ref WOD  (WOD+ ΔW )NODNF (11) where ΔW is the increase in OD finger width due to STI TCR, and NF, NOD, and WODare defined in Fig. 1.

In this paper, the standard device with WF = WOD= 2 μm

is adopted as the reference, i.e., Wref= 2 μm to minimize the

transverse stress σ and its impact on mobility. The derived model given by (9) and (11) can accurately predict μeffand Gm

measured from multifinger devices with various OD layouts, such as standard, narrow-OD, and multi-OD.

Fig. 8(a) and (b) shows that simultaneous best fitting to

μeff and Gm for multi-OD NMOS can be realized under the

condition of ΔW = 43 nm and k = 0.2888. Note that ΔW is dependent on the OD layout and varies between the stan-dard and multi-OD NMOSs. For a stanstan-dard device with more gradient trench profile, ΔW = 24 nm appears to be smaller than that of multi-OD devices. For narrow-OD devices, even with minimum OD width in W05N64, its OD width (WOD= WF = 0.5 μm) is four times larger than that of OD16 (WOD=

0.125 μm), and ΔW effect is not strong enough to overcome

μeff degradation from STI stress. It can be understood that the

ΔW ratio calculated by ΔW _ratio =ΔW × NOD× NF Weff = ΔW WOD+ ΔW (12) WOD= WF NOD (13) is about 10.6% for narrow-OD device W05N64 with WOD= WF = 0.5 μm, whereas that of multi-OD device OD16 with

WOD= 0.125 μm is significantly increased to 32.1%, which

is about three times larger than that of narrow-OD device W05N64.

(5)

Fig. 8. (a) Measuredμeffand the fitting by model (9). (b) MeasuredGmand the fitting by model (11). The simultaneous best fitting for multi-OD NMOS was realized under the condition ofΔW = 43 nm and k = 0.2888.

Fig. 9. Interface trap densityNitof W2N16, W1N32, and W05N64 devices

extracted by usingWeff, which is measured atVD= VS= VB= 0 V and

VG= Vamp= 1.2 V.

Fig. 10. Interface trap densityNitof OD1 and OD16 devices extracted by

usingWeffandW , measured at VD= VS= VB= 0 V and VG= Vamp=

1.2 V.Weff = (WOD+ ΔW )NOD∗ NFW = WODNOD∗ NF.

B. Interface Traps of Narrow-OD and Multi-OD NMOS

Strain effect on interface traps and the resulting impact on carrier mobility is one more critical concern for the deployment of strain engineering in the state-of-the-art process. Figs. 9 and 10 present the interface trap density Nit extracted by the CP

current method, according to

Icp_max= qfWeffLNit (14)

for narrow-OD and multi-OD devices, respectively, where

Icp,maxis the maximum of the measured CP current, f is the

frequency of the applied pulse waveform, and L is the gate length of the device under test. Note that the effective width

Weff is used to reflect the ΔW effect.

As shown in Fig. 9, the Nit extracted from narrow-OD

NMOS indicates nearly the same distribution for W2N16 and W1N32, and a minor decrease in W05N64 with minimal

Fig. 11. LFN SID/IDS2 measured from the standard NMOS W2N16

and narrow-OD NMOS W05N64, under the biases of VDS= 50 mV and

VGT= 0.5 V.

WF(= WOD) in this category. The results just disapprove the

conventional consideration that the narrower OD width may lead to higher Nit due to an increase in STI stress σ⊥. As for

multi-OD NMOS in Fig. 10, OD16 with minimal OD width (WOD= 0.125 μm) reveals apparently larger Nit, compared

with the standard one (OD1). The result suggests that extremely narrow OD combined with minimal OD-to-OD space in multi-OD devices may introduce a significant increase in Nitdue to a

substantially steeper trench profile, compared with narrow-OD devices. The extra trap density in the sidewall also contributes to the higher Nitof OD16. Note that ΔW should be considered

to accurately determine the effective active area and Nit in

devices with extremely narrow WOD. The increase in Nit of

OD16, which is normalized to that of OD1 and denoted as ΔNit(OD16,OD1)/Nit(OD1), is as large as 55% when ΔW is

neglected and reduced to near 10% when ΔW is considered and Weffis taken.

C. LFNSID/IDS2 in Narrow-OD and Multi-OD NMOS

Fig. 11 presents LFN in terms of SID/IDS2 measured from

narrow-OD NMOS with minimum OD width, i.e., W05N64 (WOD= 0.5 μm) and the comparison with the standard one,

i.e., W2N16 (WOD= 2 μm). Similarly, Fig. 12 makes a

com-parison of SID/IDS2 between multi-OD NMOS with minimal

OD width, i.e., OD16 (WOD= 0.125 μm) and the standard

one (OD1, WOD= 2 μm). Note that the noise spectra for all

of the devices (standard, narrow-OD, and multi-OD) follow 1/f characteristics over a wide frequency domain from 10 Hz to 10 kHz. The measured LFN manifests itself as a typical flicker noise.

As shown in Fig. 11, the narrow-OD NMOS W05N64 presents lower SID/IDS2 than the standard device W2N16.

The result looks consistent with the lower Nit for narrower

OD width, as shown in Fig. 9. More rigorously, the larger

Weff for narrower OD due to the ΔW effect have to be

considered and will be discussed later. As for multi-OD de-vices shown in Fig. 12, it is interesting to note that OD16 with extremely narrow WOD to 0.125 μm presents lower SID/IDS2 than the standard one (OD1, WOD= 2 μm)), even

though the Nitof OD16 is somewhat higher than that of OD1

(6)

Fig. 12. LFN SID/IDS2 measured from the standard NMOS OD1

(NOD= 1, WOD= 2 μm) and multi-OD NMOS OD16 (NOD=

16, WOD= 0.125 μm) under the biases of VDS= 50 mV and VGT= 0.5 V.

Fig. 13. MeasuredSID/IDS2 versus IDS underVGT= 50 mV and

vari-ousVGT (0.1–0.7 V) for standard NMOS W2N16 and narrow-OD NMOS

W05N64.

narrow-OD and multi-OD NMOS is in contradiction with the general expectation and suggests that ΔW from STI and its effect on Weff appears to be an important factor in explaining

the extraordinary results.

To further explore the mechanism responsible for LFN, the

SID/I2

DS measured at 50 Hz and under various VGT’s are

plotted versus IDS for narrow-OD and multi-OD NMOS, as

shown in Figs. 13 and 14, respectively. Note that LFN data have been collected from ten devices for each layout to perform a statistical analysis. Herein, the measured SID/IDS2 follows

a function proportional to 1/I2

DS over the whole range of

bias conditions (VGT= 0.1−0.7 V), which indicates that the

number fluctuation model given by

SID I2 DS = q2kBT λNit WeffLCox2 1 f G2 m I2 DS (15) is the dominant mechanism governing NMOS devices’ LFN, where Nit is the density of interface traps at quasi-Fermi

level [16].

The proposed number fluctuation model suggests that the

SID/IDS2 of NMOS is proportional to Nit/Weff and predicts

the decrease in LFN with increasing effective width Weff. It

was expected that the increase in compressive transverse stress

σ as well as interface traps Nit suffered by OD16 device

might aggravate interface scattering effect and increase the flicker noise [17]. However, the larger Weff may compensate or

even overcome these effects. The aforementioned mechanism

Fig. 14. MeasuredSID/IDS2 versusIDSunderVGT= 50 mV and various

VGT(0.1–0.7 V) for standard NMOS OD1(NOD= 1, WOD= 2 μm) and

multi-OD NMOS OD16(NOD= 16, WOD= 0.125 μm).

Fig. 15. Statistical distribution of LFNSID/IDS2 measured from ten dies for

multi-OD and narrow-OD NMOSs.

can explain why the multi-OD devices with extremely narrow

WOD, such as OD16, can have lower LFN, compared with OD1. Fig. 15 indicates the statistical distribution of SID/IDS2

measured from ten dies for narrow-OD and multi-OD NMOS devices to further investigate OD width scaling effect on LFN. The ratio of OD16:OD1 is in the range of 40%–87%. According to (15), the ΔW effect can lead to 32% increase in Weff for

OD16 and then 24% lower LFN, i.e., OD16 : OD1 = 76% from ΔW effect alone. Considering the statistical distribution of Nit

shown in Fig. 10, the increase in Weff of OD16 leads to a

decrease in Nit, and the ratio of OD16:OD1 is about 1.2 for the

mean value and 1.13 for the maximum in the distribution. The combined effect from ΔW and Nitindeed leads to 10%–15%

lower LFN in OD16 than OD1. The statistical data suggest that the ΔW effect may not be the only factor but is the primary factor contributing to lower LFN.

For narrow-OD devices, the Nit is similar between W2N16

and W05N64 (Fig. 8), and the larger Weff in W05N64 can

significantly overwhelm Nit and the stress effects. As a

re-sult, the smaller OD width in W05N64 leads to lower LFN (SID/IDS2 ), compared with W2N16. The investigation of layout

effects on LFN in multifinger MOSFETs discloses an interest-ing and important phenomenon that STI stress collaboratinterest-ing with STI TCR-induced ΔW constitutes the primary mechanism responsible for the OD width scaling effect on Gmand LFN.

D. High-Frequency Performance in Narrow-OD and Multi-OD NMOSs

The potential impact from the aforementioned STI stress and TCR-induced ΔW on high-frequency performance is of one

(7)

Fig. 16. Cutoff frequencyfTversusVGTmeasured from narrow-OD NMOS underVDS= 1.2 V and varying VGT.fT is extracted from the unit current

gain|H21| = 1.

Fig. 17. Cutoff frequencyfTversusVGTmeasured from multi-OD NMOS

underVDS= 1.2 V and varying VGT.fT is extracted from unit current gain

|H21| = 1.

more special concern for multifinger MOSFETs in RF circuit design. Fig. 16 illustrates the cutoff frequency fT measured

from standard (W2N16) and narrow-OD (W1N32, W05N64) NMOS under varying VGT. Note that H-parameters were

con-verted from S-parameters after two-step de-embedding (open and short), and then, fT was extracted from the extrapolation

of|H21| to unity current gain. The experimental results reveal

that narrow-OD devices suffer significant degradation of fT and

the peak fT drops from 116 GHz for W2N16 to 78 GHz for

W05N64. Referring to fT = Gm  C2 gg− Cgd2 (16)

which is an analytical model for calculating fT [18], it is

pre-dicted that fT degradation may be originated from degradation

of Gmor increase in Cgg. For narrow-OD devices, the smallest Gmappearing in W05N64 (Fig. 6) suggests one of the factors

responsible for the worst fT. Furthermore, Cggmeasured from

narrow-OD NMOS indicates the largest one for W05N64 (not shown). The collective effect from Gm and Cgg can explain fT degradation in narrow-OD devices. Regarding OD width

scaling effect on fT in multi-OD NMOS, Fig. 17 demonstrates

that OD16 with the smallest OD width (WOD= 0.125 μm)

reveals the worst fT. The peak fTdrops from 101 GHz for OD1 to 75 GHz for OD16. Cgg measured from multi-OD NMOS

indicates that OD16 suffers about 20% larger Cgg than that

of OD1 (not shown). The collective effect from smaller Gm

(Fig. 7) and larger Cggis responsible for fT degradation.

Fig. 18. fMAXversusVGTmeasured from narrow-OD NMOS underVDS=

1.2 V and varyingVGT.fMAXis determined by the unilateral gain method.

Fig. 19. fMAXversusVGTmeasured from multi-OD NMOS underVDS=

1.2 V and varyingVGT.fMAXis determined by the unilateral gain method.

The maximum oscillation frequency fMAX is another

im-portant performance parameter for RF circuit design, particu-larly for power amplifiers. The impact of layout variation on

fMAX in multifinger MOSFET is investigated as follows: In

this paper, fMAX was determined by the conventionally used

unilateral gain method. Fig. 18 presents fMAXextracted from

standard and narrow-OD NMOS. It reveals a very interest-ing result that W1N32 attains about 10 GHz higher fMAX

than the standard (W2N16), even though W1N32 suffers the lower fT than W2N16 (Fig. 16). However, W05N64 with the lowest fT remains the worst in terms of fMAX. Through an

equivalent circuit analysis on unilateral gain (U), fMAXcan be

approximated by

fMAX =

fT

2Rg(gDS+ 2πfTCgd) + gDS(Ri+ Rs)

(17) in which Rgand fT appear as two key parameters controlling

fMAX [11]. This expression predicts that the higher fT and

lower Rg can increase fMAX. For narrow-OD devices, the

smaller WF(= WOD) and larger NFplay a multiplied effect to

reduce Rg(not shown). The higher fMAXachieved by W1N32,

compared with W2N16, suggests that the significant reduction in Rg can compensate or even overcome the impact from fT

degradation. However, the advantage of Rgsuppression cannot

be obtained by multi-OD devices due to the feature of fixed

NF [Fig. 1(c)]. The fMAX measured from multi-OD NMOS,

as shown in Fig. 19, proves the expectation that the trend of

fMAXsimply follows that of fT (Fig. 17), i.e., the smaller WOD

leads to the lower fMAX and OD16 suffers worse degradation

(8)

an abnormal Gmincrease for extremely narrow OD width to

WOD= 0.125 μm. The observed results suggest that STI stress

is not the only mechanism governing the electrical property in miniaturized devices. STI TCR-induced ΔW is identified as another key factor, which may overcome STI stress effect in determining channel current and Gm. Semi-empirical formulas

have been derived to successfully predict WOD scaling effect

on μeff and Gm. Taking this method, ΔW can be precisely

extracted based on a simultaneous best fitting to μeff and Gm,

and the resulting increase in effective width (Weff) is

dramati-cally large to about 34% for OD16 with WOD= 0.125 μm. The

larger Weff becomes, the more it contributes to reducing LFN

and overcoming Niteffect in narrow-OD and multi-OD devices

with sufficiently small WOD. The reduction in LFN with OD

width scaling is the other evidence reflecting STI TCR-induced ΔW effect.

Unfortunately, the OD width scaling leads to a negative impact on high-frequency performance like fT and fMAX, due

to Gmdegradation and undesired increase in Cgg. An improved

open de-embedding method can reduce the parasitic capac-itances from intermetal coupling but cannot eliminate gate-related fringing capacitances. The multifinger MOSFETs with miniaturized OD width cannot prevent from fT degradation.

The tradeoff between LFN and high-frequency performance identified from this work provides an important layout guide-line for analog and RF circuit design.

ACKNOWLEDGMENT

The authors would like to thank the National Device Labo-ratory (NDL) for their support during noise measurement and the Chip Implementation Center (CiC) for test key tape-out and device fabrication.

REFERENCES

[1] J. Xue, Z. Ye, Y. Deng, H. Wang, L. Yang, and Z. Yu, “Layout-dependent STI stress analysis and stress-aware RF/analog circuit design optimization,” in Proc. ACM/IEEE Int. Conf. Comput.-Aided Des., 2009, pp. 521–528 .

[2] K.-W. Su, Y.-M. Sheu, C.-K. Lin, S.-J. Yang, W.-J. Liang, X. Xi, C.-S. Chiang, J.-K. Her, Y.-T. Chia, C. H. Dim, and C. Hu, “A scalable model for STI mechanical stress effect on layout dependence of MOS electrical characteristics,” in Proc. IEEE Custom Integr. Circuits Conf., 2003, pp. 245–248.

[3] A. B. Kahng, P. Sharma, and R. O. Topaloglu, “Chip optimization through STI-stress-aware placement perturbations and fill insertion,” IEEE Trans.

Comput.-Aided Design Integr. Circuits Syst., vol. 27, no. 7, pp. 1241–

1252, Jul. 2008.

[4] R. Li, L. Yu, H. Xin, Y. Dong, K. Tao, and C. Wang, “A comprehensive study of reducing the STI mechanical stress effect on channel-width-dependentIdsat,” Semicond. Sci. Technol., vol. 22, no. 12, pp. 1292–

1297, Nov. 2007.

influence,” IEEE Trans. Electron Devices, vol. 49, no. 4, pp. 605–612, Apr. 2002.

[9] C.-Y. Chan, Y.-S. Lin, Y.-C. Huang, S. S. H. Hsu, and Y.-Z. Juang, “Edge-extended design for improved flicker noise characteristics in 0.13-μm RF NMOS,” in Proc. IEEE MTT-S Int. Microw. Symp., 2007, pp. 441–444. [10] Y.-L.R. Lu, Y.-C. Liao, W. McMahon, Y.-H. Lee, H. Kung, R. Fastow,

and S. Ma, “The role of shallow trench isolation on channel width noise scaling for narrow width CMOS and flash cells,” in Proc. Symp. VLSI

Technol. Syst. Appl., 2008, pp. 85–86.

[11] H. Shimomura, A. Matsuzawa, H. Kimura, G. Hayashi, T. Hirai, and A. Kanda, “A mesh-arrayed MOSFET (MA-MOS) for high-frequency analog applications,” in VLSI Symp. Tech. Dig., 1997, pp. 73–74. [12] A. B. M Elliot, “The use of charge pumping currents to measure surface

state densities in MOS transistors,” Solid State Electron., vol. 19, no. 3, pp. 214–247, Mar. 1976.

[13] M. Miyamoto, H. Ohta, Y. Kumagai, Y. Sonobe, K. Ishibashi, and Y. Tainaka, “Impact of reducing STI-induced stress on layout dependence of MOSFET characteristics,” IEEE Trans. Electron Devices, vol. 51, no. 3, pp. 440–443, Mar. 2004.

[14] A. Ono, R. Ueno, and I. Sakai, “TED control technology for suppression of reverse narrow channel effect in 0.1μm MOS devices,” in IEDM Tech.

Dig., 1997, pp. 227–230.

[15] G. Scott, J. Lutze, M. Rubin, F. Nouri, and M. Manley, “NMOS drive current reduction caused by transistor layout and trench isolation induced stress,” in IEDM Tech. Dig., 1999, pp. 827–830.

[16] G. Ghibaudo, O. Roux, C. H. Nguyen-Duc, F. Balestra, and J. Brini, “Im-proved analysis of low frequency noise in field-effect MOS transistors,”

Phys. Stat. Sol. (A), vol. 124, no. 2, pp. 571–581, Apr. 1991.

[17] S. Maeda, Y. S. Jin, J. A. Choi, S. Y. Oh, H. W. Lee, J. Y. Woo, M. C. Sun, J. H. Ku, K. Lee, S. G. Bae, S. G. kang, J. H. Yang, Y. W. Kim, and K. P. Suh, “Impact of mechanical stress engineering on flicker noise characteristics,” in VLSI Symp. Tech. Dig., 2004, pp. 102–103. [18] T. Manku, “Microwave CMOS—Device physics and design,” IEEE J.

Solid-State Circuits, vol. 34, no. 1, pp. 277–285, Mar. 1999.

Kuo-Liang Yeh (M’09) received the B.S.E.E.

de-gree from National Chiao-Tung University (NCTU), Hsinchu, Taiwan, in 1995 and the M.S.E.E. degree from National Taiwan University, Taipei, Taiwan, in 1997. He is currently working toward the Ph.D. degree in electronics engineering in the Institute of Electronics Engineering, NCTU.

In 1999, he joined Taiwan Semiconductor Man-ufactory Company Inc., Hsinchu, where he has worked on process integration and yield improve-ment. From 2004 to 2007, he has been a Senior Engineer with MediaTek Inc., Hsinchu. He is currently a Senior Manager with the Silicon Motion Technology Corporation, Hsinchu. He has authored more than ten technical publications in international journals and conference proceedings. His research interests include the characterization and parameter extraction of CMOS devices for modeling and circuit simulation, as well as the protection of intellectual property rights.

(9)

device design and VLSI technology development. In 1984, she joined the ERSO/ITRI, where she had been engaged in semiconductor integrated circuit technologies with a broad scope that covers high-voltage, high-power, submi-crometer project, and high-speed SRAM technologies. From 1994 to 1998, she was with Macronix International Corporation and engaged in high-density and

since 2008. She has authored or coauthored more than 60 technical papers. She is the holder of 19 U.S. patents in her professional field. Her current research interests include RF/MS CMOS device design and modeling for low-power and low-noise nanoscale CMOS noise modeling and strain engineering effects, broadband and scalable inductors modeling, novel nonvolatile memory technologies, and device integration technologies for system-on-chip.

數據

Fig. 2. LFN measurement system setup consisting of an Agilent dynamic signal analyzer (DSA 35670), low noise amplifier (LNA SR570), and Agilent 4156B for dc power supply
Fig. 2. LFN measurement system setup consisting of an Agilent dynamic signal analyzer (DSA 35670), low noise amplifier (LNA SR570), and Agilent 4156B for dc power supply p.2
Fig. 3. (a) CP measurement system consisting of dc source Agilent 4156B, pulse generator 81110A, and switching matrix Keithley 707A
Fig. 3. (a) CP measurement system consisting of dc source Agilent 4156B, pulse generator 81110A, and switching matrix Keithley 707A p.2
Fig. 5. Linear VT versus W OD for narrow-OD and multi-OD NMOSs under
Fig. 5. Linear VT versus W OD for narrow-OD and multi-OD NMOSs under p.3
Fig. 7. (a) Transconductance Gm versus V GT . (b) Maximum transconduc-
Fig. 7. (a) Transconductance Gm versus V GT . (b) Maximum transconduc- p.3
Fig. 6. (a) Transconductance Gm versus V GT . (b) Maximum transcon-
Fig. 6. (a) Transconductance Gm versus V GT . (b) Maximum transcon- p.3
Fig. 4. Schematic of STI stresses along the longitudinal and transverse directions, which are defined as σ// and σ⊥ in MOSFETs with three different layouts
Fig. 4. Schematic of STI stresses along the longitudinal and transverse directions, which are defined as σ// and σ⊥ in MOSFETs with three different layouts p.3
Fig. 8. (a) Measured μ eff and the fitting by model (9). (b) Measured Gm and the fitting by model (11)
Fig. 8. (a) Measured μ eff and the fitting by model (9). (b) Measured Gm and the fitting by model (11) p.5
Fig. 11. LFN S ID /I DS 2 measured from the standard NMOS W2N16
Fig. 11. LFN S ID /I DS 2 measured from the standard NMOS W2N16 p.5
Fig. 9. Interface trap density N it of W2N16, W1N32, and W05N64 devices
Fig. 9. Interface trap density N it of W2N16, W1N32, and W05N64 devices p.5
Fig. 10. Interface trap density N it of OD1 and OD16 devices extracted by
Fig. 10. Interface trap density N it of OD1 and OD16 devices extracted by p.5
Fig. 12. LFN S ID /I DS 2 measured from the standard NMOS OD1
Fig. 12. LFN S ID /I DS 2 measured from the standard NMOS OD1 p.6
Fig. 14. Measured S ID /I DS 2 versus I DS under V GT = 50 mV and various
Fig. 14. Measured S ID /I DS 2 versus I DS under V GT = 50 mV and various p.6
Fig. 15. Statistical distribution of LFN S ID /I DS 2 measured from ten dies for
Fig. 15. Statistical distribution of LFN S ID /I DS 2 measured from ten dies for p.6
Fig. 13. Measured S ID /I DS 2 versus I DS under V GT = 50 mV and vari-
Fig. 13. Measured S ID /I DS 2 versus I DS under V GT = 50 mV and vari- p.6
Fig. 18. f MAX versus V GT measured from narrow-OD NMOS under V DS =
Fig. 18. f MAX versus V GT measured from narrow-OD NMOS under V DS = p.7
Fig. 19. f MAX versus V GT measured from multi-OD NMOS under V DS =
Fig. 19. f MAX versus V GT measured from multi-OD NMOS under V DS = p.7
Fig. 17. Cutoff frequency fT versus V GT measured from multi-OD NMOS
Fig. 17. Cutoff frequency fT versus V GT measured from multi-OD NMOS p.7
Fig. 16. Cutoff frequency fT versus V GT measured from narrow-OD NMOS under V DS = 1.2 V and varying V GT
Fig. 16. Cutoff frequency fT versus V GT measured from narrow-OD NMOS under V DS = 1.2 V and varying V GT p.7

參考文獻

Updating...