Silicon Oxides: SiO 2

Transcription

1 Silicon Oxides: SiO 2 Uses: diffusion masks surface passivation gate insulator (MOSFET) isolation, insulation Formation: grown / native thermal: highest quality anodization deposited: C V D, evaporate, sputter vitreous silica: material is a GLASS under normal circumstances can also find crystal quartz in nature m.p C; glass is unstable below 1710 C BUT devitrification rate (i.e. crystallization) below 1000 C negligible bridging oxygen nonbridging oxygen silicon network modifier network former hydroxyl group Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

2 Growth of SiO 2 from Si in dry (<< 20 ppm H2O) oxygen Si O 2 SiO 2 once an oxide is formed, how does this chemical reaction continue? does the oxygen go in or the silicon go out? density / formula differences ρ SiO2 = 2.25 gm/cm 3, GMW = 60 ρ Si = 2.3 gm/cm 3, GMW = 28 oxide d thick consumes d SiO 2 a layer 0.44d thick of Si original silicon surface 0.44d bare silicon in air is always covered with about 1520 Å of oxide, upper limit of ~ 40 Å it is possible to prepare a hydrogen terminated Si surface to retard this native oxide formation Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

3 Wet oxidation of Si overall reaction is Si 2 H 2 O SiO 2 H 2 proposed process H 2 O SiOSi SiOH SiOH diffusion of hydroxyl complex to SiO 2 Si interface Si OH Si O H Si O Si Si Si H 2 Si O Si this results in a more open oxide, with lower density, weaker structure, than dry oxide ρ wet gm / cm 3 Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

4 Oxide growth kinetics basic model is the Grove and Deal Model supply of oxidizer is limited by diffusion through oxide to growth interface Fick s First Law: flux j = D N oxidizer x concentration N 0 N1 moving growth interface gas oxide silicon simplest approximation: x N x = N 0 N 1 x Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

5 Oxidizer concentration gradient and flux concentration N 0 N1 moving growth interface gas oxide silicon x Noxidizer N0 N1 j = D D x x N 0 is limited by the solid solubility limit of the oxidizer in the oxide! N 0 O2 ~ 5 x cm 1000 C N 0 H2O ~ 3 x cm 1000 C flux of oxidizer j at SiO 2 / Si interface consumed to form new oxide j' = k N 1 k is the chemical reaction rate constant in steady state, flux in must equal flux consumed steady state j' = j k N 1 = D N 0 N 1 x solve for N1, sub back into flux eq j = D N 0 x D k Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

6 Relation between flux and interface position flux: #oxidizer molecules crossing interface per unit area per unit time # cm 2 sec 1 rate of change of interface position: dx / dt (interface velocity) cm sec 1 n: # of oxidizer molecules per unit volume of oxide: n = ρ SiO 2 N A GMW SiO2 # cm 3 then relation is just d x dt 2 for H 2 O 1 for O 2 = j n = = cm 3 2 for H 2 O 1 for O 2 DN 0 n x Dk now integrate with appropriate initial condition Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

7 Grove and Deal relation setting 2D/k = A function of what s diffusing, what it s diffusing in, and what it reacts with 2DN 0 /n = B function of what s diffusing and what it s diffusing in initial condition x (t = 0) = x i integration gives A t τ x() t = A 4B LiveMath where τ represents an offset time to account for any oxide present at t = 0 τ = x i ( ) 2 A x i B Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

8 Limiting behavior of Grove & Deal oxidation model A t τ () t = A 4B x 2 short times t τ << A 2 4 B x() t = A 2 1 t τ A 2 4B 1 x() t A t τ A 2 1 4B = B A ( t τ) thickness is linearly increasing with time characteristic of a reaction rate limited process B/A is the linear rate constant B 0 0 A 2 D N = n linear rate constant depends on 2 D k = N k n reaction rate between oxidizer and silicon (k) AND solid solubility of oxidizer in oxide (N 0 ) temperature dependence mainly from reaction rate Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

9 Rate constants and Arrhenius plots E A = 0.5eV thermally activated process i.e., process must thermally y overcome an energy barrier E A = 1eV y = y o e E A kt LiveMath x x x temperature plot log(y) vs 1/T if process has the simple thermallyactivated behavior you will get a straight line! 0 E A = 0.5eV log[y] 0 E A = 1eV /[temperature] Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

10 x Rate constant behavior: linear rate constant A t τ () t = A 4B x 2 B A () t ( tτ) linear rate constant depends on AND B A = 2DN 0 n reaction rate between oxidizer and silicon (k) is orientation dependent temperature dependence mainly from energy required to break SiSi bond solid solubility of oxidizer in oxide (N 0 ) t τ << A 2 4 B 2D k = N 0 k n (111) Si (100) Si H20 (640 Torr) EA = 2.05 ev (111) Si (100) Si 103 dry O2 EA = 2.0 ev adapted from Ghandhi Temperature 1000/T (K1) Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin Linear rate constant B/A (µm/hour) plane 1200 C1000 C 800 C available bonds (100) 6.8 x10 14 cm 2 (111) 11.7 x10 14 cm 2 B A ( 111) B A ( 100 ) = 1.68 ratio 1.7 (from Sze, 2nd ed., p. 110)

11 Limiting behavior of Grove & Deal oxidation model A t τ x() t = A 4B long times t τ >> A 2 4 B x() t = A 2 1 t τ A 2 4B 1 x() t A 2 t τ A 2 4B = B ( t τ) dependence is parabolic : (thickness) 2 time characteristic of a diffusion limited process B is the parabolic rate constant B = 2 D N 0 n parabolic rate constant depends on diffusivity of oxidizer in oxide (D) AND solid solubility of oxidizer in oxide (N 0 ) temperature dependence mainly from diffusivity Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

12 Rate constant behavior: parabolic rate constant x () t B ( tτ) B = 2 D N 0 n parabolic rate constant depends on diffusivity of oxidizer in oxide (D) and solid solubility of oxidizer in oxide (N 0 ) Solid 1000 C H 2 O 3 x cm 3 O 2 5 x cm 3 Parabolic rate constant B (µm 2 / hour) C 1100 C 1000 C 900 C 800 C 700 C 600 C H 2 0 (76 0 T orr ) E A = ev dr y O 2 E A = ev is NOT orientation dependent Temperatur e 1000/T ( K 1 ) IS oxidizer dependent temperature dependence mainly from diffusivity of oxidizer in oxide adapted from Ghandhi Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

13 boron Effect of Si doping on oxidation kinetics k = C ox / C Si ~ 3 dopants accumulate in oxide little effect on linear rate constant B/A ( = N o k/ n) can increase parabolic rate constant B ( = 2DN o / n ) really only significant for N boron > ~10 20 cm 3 phosphorus k = C ox / C Si ~ 0.1 dopants pileup at silicon surface little effect on parabolic rate constant B increases linear rate constant B/A again, really only significant for N phosphorus > ~10 20 cm 3 concentration concentration k > 1, slow SiO 2 diffuser oxide oxide silicon k < 1, slow SiO 2 diffuser silicon Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

14 Oxidation thicknesses Oxide thickness (microns) dry oxidation (100) silicon dry 1200 C 1000 C 1100 C 900 C Time (hours) wet oxidation Oxide thickness (microns) Torr partial pressure is typical (vapor pressure over liquid 95 C) (100) silicon steam 1050 C 1000 C 950 C C 800 C 1100 C 1150 C Time (hours) Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

15 Pressure Effects on Oxidation grow thick oxides at reduced time / temperature product use elevated pressures to increase concentration of oxidizer in oxide for steam, both B and B/A ~ linear with pressure rule of thumb: constant growth rate, if for each increase of 1 atm pressure, temperature is reduced ~ 30 C. pressures up to 25 atm have been used (commercial systems: HiPOx, FOX) Oxide Thickness ( microns) Pyrogenic steam 900 º C (111) (100) 20 atm 10 atm 5 atm 1 atm adapted from Sze, 2nd, p Time (hour) Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

16 Oxidation isolation how do you insulate (isolate) one device from another? what about using an oxide? try growing it first since it ll be long, high temp now etch clear regions to build devices in oxide devices silicon oxidation, etch it doesn t do any good!! Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

17 Masking of oxidation and isolation techniques would like to form thick oxide between device regions with minimum step heights mask oxidation using material with low water diffusivity / solubility: Local Oxidation of Silicon (LOCOS) process Si 3 N 4 (silicon nitride) pad induces high stress, must place pad layer below to prevent dislocations for oxide pad layer, t pad ~ 0.25 t nitride do have lateral diffusion at mask edge, produces bird s beak lateral encroachment oxide thickness can reduce using more complex pads SiO 2 / poly / nitride (~15nm/50nm/150nm) helps nitride mask silicon SiO 2 bird s beak before oxidation after oxidation Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

18 Use of SiO 2 as a diffusion mask Oxide Mask Thickness (microns) mask against boron, phosphorus, arsenic diffusion most impurities transported to slice surface as an oxide P 2 O 5, B 2 O 3 reacts with oxide to form mixed phosphosilcate (or borosilicate) glass reaction continues until full thickness of masking oxide is converted doping of underlying Si commences adapted from Nicollian and Brews, p. 732 phosphorus boron Time (minutes) 1200 C 1100 C 1000 C 900 C 1200 C 1100 C 1000 C 900 C simple parabolic relationship: phosphorus: boron: Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin ( x m ) 2 t m ( x m ) 2 t m = eV e sec kt cm 2 = eV kt cm 2 e sec

19 Other problems in oxidation: MOS threshold stability for an MIS structure if ionized impurities are in the insulator, what happens? if the ionized impurities are close to the metal, they are screened, and the silicon surface remains unchanged threshold voltage of MOS device is critically dependent on the location and amount of mobile ionic contamination in SiO 2 gate insulator stability can be adversely affected Metal Oxide Silicon image charge fixed surface states mobile ionized impurities in oxide (sodium), Qm image charge Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

20 BiasTemperatureStress (BTS) technique apply field of approximately 1MV/cm between capacitor metallization and substrate heat sample to ~200 o C to increase diffusion rate of Na allows evaluation of contamination levels E 1MV/cm T 200 C Metal Oxide Silicon mobile ionized impurities in oxide (sodium), Q m Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

21 BTS impurity drift apply field of approximately 1MV/cm across gate insulator heat sample to ~200 o C to increase diffusion rate of Na most of the impurities have drifted to near the silicon surface screening now due to electrons at silicon surface can cause surface inversion!! process is reversible! Metal Oxide Silicon image charge external electric field mobile ionized impurities in oxide (sodium), Q m image charge fixed surface states Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

22 CV measurements Cox capacitance ptype substrate after BTS Q m / Cox before BTS C T voltage (metal wrt substrate) measure CV curves before and after BTS Q m C ox x V t ρ Q m / t ox A cap q this occurs even at room temperature! threshold voltage shift V t can be MANY volts! Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

23 Other problems in oxidation: oxidation induced stacking faults typically ~95% of all stacking faults are OSFs essentially an extra (111) plane density can vary from ~0 to 10 7 / cm 2 highly dependent on process (high temperature) history size (length) at surface can be many microns growth is related to presence of excess unoxidized Si at SiSiO 2 interface heterogeneous coalescence of excess Si on nucleation centers produce OSFs any process that produces an excess of Si vacancies will inhibit OSF formation/growth Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

24 Chlorine Gettering in SiO 2 effects of chlorine during oxidation reduction of OSFs increased dielectric breakdown strength improved minority carrier lifetimes improved MOS threshold stability when oxide is grown in presence of chlorine the Cl getters ionic contaminants such as Na Cl in gas stream reacts with Na diffusing from the furnace walls Cl is incorporated into the grown oxide near the silicon interface (~20nm from Si) and can capture mobile sodium preventing threshhold instabilities efficiency of chlorine gettering can be evaluated by alternately drifting the Na back and forth from metal to silicon side of the test capacitor only effective for oxidation temperature above ~1100 C Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

25 Restrictions on the use of Cl Gettering chlorine injection techniques mix HCl in O 2 gas stream must be VERY dry or severe corrosion problems can occur produces large quantities of H 2 O in furnace: 4HCl O 2 2Cl 2 2H 2 O mix trichloroethylene (TCE) / trichloroethane (TCA) in O 2 gas stream produces less water: C 2 HCl 3 2O 2 HCl Cl 2 2CO 2 for TCE must have large excess of O 2 present to prevent carbon deposits under low temperatures and low oxygen conditions can form phosgene COCl 2 only effective if oxidation temperature 1100 C large Cl concentrations, very high temperatures, or very long oxidation times can cause rough oxides very high Cl can cause blisters and separation of SiO 2 from Si Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

26 Minimum feature (microns) Gate Oxide thickness trends K 4K 16K 64K 256K 1M 4M 16M 64M 256M 1G 16G year (dram: intro or production) MOS oxide thickness scales with gate length circa 1998 thickness was ~ 4nm (64M and 256M DRAM technology) Oxide thickness (Angstroms) Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

27 Oxidation thicknesses: Grove & Deal calculations wet oxidation oxide thickness (microns) 640 Torr partial pressure fits very well to measurement (100) Si 1100ºC 1000ºC 900ºC wet oxidation time (hours) calculation Oxide thickness (microns) Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin (100) silicon steam 1050 C 1000 C 950 C C 800 C 1100 C 1150 C Time (hours) measurement

28 Problems with the simple Grove/Deal model of oxidation Arrhenius plots of linear and parabolic rate constants are not straight at low temperatures (T < 900 o C). dry oxidation growth curves do not extrapolate back to zero oxide thickness at zero time Wet T ( C) A (µm) B (µm 2 /hr) B/A (µm/hr) τ (hr) x i (nm) ( Torr H 2 O) Dry (760 Torr O 2 ) ?? ~2.6 x time (hours) must use nonphysical boundary condition of either an offset time τ or x i 200 Å in order to fit data Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin Dry Oxidation Thickness (µm) (100) Si 1000ºC 1100ºC 900ºC

29 Possible models to explain rapid initial growth micropores and intrinsic stress in low temperature thin oxides micropores: ~10 Å about 100 Å apart can visualize as small irregularities that mask oxidation, adjacent regions grow up around "holes" diffusion of O 2 down pores can fit rapid initial growth stage and curvature of Arrhenius plots Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

30 FieldEnhanced Diffusion experiments with external applied fields imply O 2 is charged during oxidation: O 2 O 2 h because of mobility differences have ambipolar diffusion effects: range of effect is approximately the Debye length λ D λ D 1/ N ~ Å for O 2 in SiO 2 ~5 Å for H 2 O in SiO 2 gas ~λ D O 2 hole oxide silicon Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

31 Thin oxide growth thin oxides can be grown controllably use reduced pressure usually dry O 2 oxidation need pressure (1 atms = 3x10 19 ) near solid solubility limit (5x C) ~ 10 3 atms ( Torr used) use low temperature!!! intrinsic stress in low temperaturegrown oxides: low temp oxides tend to have higher density not thermal expansion stress at high temp viscosity of SiO 2 low enough to allow plastic flow at low temp viscosity is too high Oxide thickness (nm) (100) silicon, dry oxygen 1000 C adapted from Sze, 2nd ed,p Oxidation time (minutes) Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin 950 C 900 C 850 C 800 C 10 3

32 ITRS roadmap (2000) requirements ( ) technology node (nm) year min gate length (nm) equivalent gate oxide thickness (nm) projections require l gate ~ 65nm, effective oxide thickness ~ 11.5nm EOT : EOT εr = t physical ε problem: excessive leakage current and boron penetration for oxide thicknesses < 1.5nm alternative high k dielectrics rsio 2 Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin

33 Diffusion Mechanisms probability of movement ν 4ν o e E kt, ν o ~ sec 1 interstitial diffusers E move ~ ev T = 300K: ν ~ 1 jump per minute substitutional impurity vacancy T = 1300K: ν ~ 10 9 jumps per sec substitutional diffusers E move ~ 3 4 ev T = 300K: ν ~ 1 jump per years! T = 1300K: ν ~ few jumps per sec interstitial impurity interstitialcy mechanism Dean P. Neikirk 1999, last update October 2, Dept. of ECE, Univ. of Texas at Austin