Level 4 — ApplicationSemiconductor Fabrication

Semiconductor Fabrication

60 minutes60 marksprintable — key stays hidden on paper

Level 4 — Application (novel problems, no hints) Time limit: 60 minutes Total marks: 60

Use ...... for inline math and ...... for display. Show all working. Physical constants: Avogadro NA=6.022×1023 mol1N_A = 6.022\times10^{23}\ \text{mol}^{-1}; silicon density 2.33 g/cm32.33\ \text{g/cm}^3; silicon molar mass 28.09 g/mol28.09\ \text{g/mol}.


Question 1 — Lithography resolution and multi-patterning (14 marks)

A fab runs an ArF immersion DUV scanner (λ=193 nm\lambda = 193\ \text{nm}) with a numerical aperture NA=1.35NA = 1.35.

(a) Using the Rayleigh criterion CD=k1λ/NACD = k_1 \lambda / NA with a single-exposure process factor k1=0.30k_1 = 0.30, compute the minimum resolvable half-pitch. (3)

(b) The target design requires a 20 nm half-pitch metal layer. Determine how many times you must divide the pitch (i.e. the multi-patterning factor NN, where achievable half-pitch =single-exposure HP/N= \text{single-exposure HP}/N) to reach the target, rounding NN up to the nearest integer power-of-two-compatible value used in LELE/SADP schemes. State whether SADP (self-aligned double patterning) alone is sufficient. (4)

(c) An EUV scanner with λ=13.5 nm\lambda = 13.5\ \text{nm} and NA=0.33NA = 0.33 is proposed as an alternative. Compute its single-exposure half-pitch at the same k1=0.30k_1 = 0.30, and state one physical reason EUV needs reflective (not transmissive) optics. (4)

(d) Give one manufacturing disadvantage of reaching 20 nm HP by multi-patterning versus by EUV single exposure. (3)


Question 2 — Thermal oxidation budget (12 marks)

A wafer is oxidized to grow a gate-quality SiO2\text{SiO}_2 layer. In the linear (thin-oxide) regime the oxide thickness grows as xox=(B/A)tx_{ox} = (B/A)\,t, where the linear rate constant B/A=0.05 μm/hrB/A = 0.05\ \mu\text{m/hr} at the process temperature.

(a) Compute the time required to grow a 4 nm oxide in the linear regime. (3)

(b) Growing SiO2\text{SiO}_2 consumes underlying silicon. Given that 0.440.44 nm of Si is consumed per 1.0 nm of oxide grown, compute the silicon thickness consumed for the 4 nm oxide, and the net surface height change relative to the original Si surface (state whether the oxide surface sits above or below the original surface). (5)

(c) Explain, in terms of the Deal–Grove model, why doubling the target oxide thickness in the parabolic (thick-oxide) regime requires roughly four times the oxidation time. (4)


Question 3 — Wafer economics, yield and binning (16 marks)

A 300 mm wafer is patterned with square dies of edge 8 mm.

(a) Estimate the maximum number of whole dies per wafer using Nπ(D/2)2SπD2SN \approx \dfrac{\pi (D/2)^2}{S} - \dfrac{\pi D}{\sqrt{2S}}, where DD is wafer diameter and SS is die area. Show both terms. (4)

(b) Using the Poisson yield model Y=eD0AY = e^{-D_0 A} with defect density D0=0.15 defects/cm2D_0 = 0.15\ \text{defects/cm}^2 and die area AA (in cm²), compute the yield and the expected number of good dies. (5)

(c) The good dies are binned by max stable clock: 30% at 3.5 GHz, 45% at 3.0 GHz, 25% at 2.5 GHz. Selling prices are $210, $150, $95 respectively. Compute expected revenue per wafer from good dies. (4)

(d) Management wants to raise yield to 0.90. Compute the required defect density D0D_0 and comment on whether this is more realistically achieved via process cleanliness or die-size reduction. (3)


Question 4 — Transistor scaling: FinFET → GAA (10 marks)

(a) A FinFET has a rectangular fin of height Hfin=42 nmH_{fin} = 42\ \text{nm} and width Wfin=8 nmW_{fin} = 8\ \text{nm}, gated on three sides. Compute the effective electrostatic channel width WeffW_{eff} per fin. (3)

(b) A gate-all-around nanosheet device replaces this fin with a stack of 3 nanosheets, each 30 nm30\ \text{nm} wide and 5 nm5\ \text{nm} thick, gated on all four sides. Compute WeffW_{eff} for the stack. (4)

(c) State one electrostatic advantage of the GAA geometry over FinFET that is not captured purely by the WeffW_{eff} number. (3)


Question 5 — Copper damascene & deposition sequence (8 marks)

(a) List, in correct process order, the main steps of a single (Cu) damascene interconnect module, from dielectric deposition to a planar Cu-filled trench. (4)

(b) Explain why electroplated Cu requires both a barrier layer and a seed layer, naming a typical technique used to deposit each. (4)


Answer keyMark scheme & solutions

Question 1 (14)

(a) CD=k1λ/NA=0.30×193/1.35=57.9/1.35=42.9 nmCD = k_1\lambda/NA = 0.30 \times 193 / 1.35 = 57.9/1.35 = 42.9\ \text{nm}. So single-exposure half-pitch 42.9 nm\approx 42.9\ \text{nm}. (3) — (1 formula, 1 substitution, 1 answer)

(b) Need HP = 20 nm. N=42.9/20=2.14N = 42.9/20 = 2.14. Round up to a multi-patterning-compatible value → N=4N = 4 (LELE gives ×2, LELELE/SADP schemes go ×2, ×4). A single SADP gives factor 2 → HP =42.9/2=21.4 nm= 42.9/2 = 21.4\ \text{nm}, which is just above 20 nm target, so SADP alone is marginal/insufficient; SAQP (×4) or LE³ is safely required. (4) — (1 for NN calc, 1 rounding, 1 SADP HP=21.4 nm, 1 conclusion)

(c) EUV: CD=0.30×13.5/0.33=4.05/0.33=12.3 nmCD = 0.30 \times 13.5 / 0.33 = 4.05/0.33 = 12.3\ \text{nm} single exposure — easily resolves 20 nm HP in one pass. EUV needs reflective optics because 13.5 nm light is strongly absorbed by all known lens materials (no transmissive glass), so mirrors (Mo/Si multilayer Bragg reflectors) and vacuum are used. (4) — (1 formula, 1 = 12.3 nm, 1 single-exposure OK, 1 absorption reason)

(d) Any one: multi-patterning multiplies cost/time (2–4× the exposures + intermediate etch/deposition), increases overlay/edge-placement error, and lowers throughput/yield versus one EUV exposure. (3)

Question 2 (12)

(a) t=xox/(B/A)=0.004 μm/0.05 μm/hr=0.08 hr=4.8 mint = x_{ox}/(B/A) = 0.004\ \mu\text{m} / 0.05\ \mu\text{m/hr} = 0.08\ \text{hr} = 4.8\ \text{min}. (3)

(b) Si consumed =0.44×4=1.76 nm= 0.44 \times 4 = 1.76\ \text{nm}. Oxide grows 4 nm total but 1.76 nm of it is below the original Si surface; the oxide top rises 41.76=2.24 nm4 - 1.76 = 2.24\ \text{nm} above the original Si surface. (5) — (2 consumed, 2 net rise, 1 direction)

(c) In parabolic regime xox2Btx_{ox}^2 \approx B t (diffusion of oxidant through existing oxide is rate-limiting). Since xoxtx_{ox}\propto\sqrt{t}, doubling xoxx_{ox} needs tt increased by 22=42^2 = 4. Physically the growing oxide layer lengthens the diffusion path for oxidant to reach the Si/SiO₂ interface, slowing growth. (4)

Question 3 (16)

(a) D=300 mmD = 300\ \text{mm}, S=64 mm2S = 64\ \text{mm}^2. Term1 =π(150)2/64=π22500/64=1104.5= \pi(150)^2/64 = \pi\cdot22500/64 = 1104.5. Term2 =π300/128=942.48/11.314=83.3= \pi\cdot300/\sqrt{128} = 942.48/11.314 = 83.3. N1104.583.3=1021 diesN \approx 1104.5 - 83.3 = 1021\ \text{dies} (≈1021). (4)

(b) A=0.64 cm2A = 0.64\ \text{cm}^2; Y=e0.15×0.64=e0.096=0.9085Y = e^{-0.15\times0.64} = e^{-0.096} = 0.9085. Good dies =1021×0.9085=927.6928= 1021 \times 0.9085 = 927.6 \approx 928. (5)

(c) Weighted price per good die =0.30(210)+0.45(150)+0.25(95)=63+67.5+23.75=154.25= 0.30(210)+0.45(150)+0.25(95) = 63+67.5+23.75 = 154.25. Revenue = 928 \times 154.25 = \143{,}144 (≈\143k). (4)

(d) 0.90=eD0(0.64)D0=ln(0.90)/0.64=0.10536/0.64=0.1646 defects/cm20.90 = e^{-D_0(0.64)} \Rightarrow D_0 = -\ln(0.90)/0.64 = 0.10536/0.64 = 0.1646\ \text{defects/cm}^2. This is above the current 0.15, so current yield already ~0.91 > 0.90 — target is already met; alternatively reducing die area also raises yield. Comment: cleanliness (lower D0D_0) and smaller dies both help; here yield already exceeds 0.90. (3)

Question 4 (10)

(a) Tri-gate: Weff=2Hfin+Wfin=2(42)+8=92 nmW_{eff} = 2H_{fin} + W_{fin} = 2(42)+8 = 92\ \text{nm}. (3)

(b) Per sheet (all-around): perimeter =2(W+t)=2(30+5)=70 nm= 2(W+t) = 2(30+5) = 70\ \text{nm}. Stack of 3: Weff=3×70=210 nmW_{eff}=3\times70 = 210\ \text{nm}. (4)

(c) GAA wraps the gate fully around the channel, giving superior electrostatic control (better subthreshold slope, reduced short-channel effects/DIBL, lower leakage) and allowing continued gate-length scaling — not captured by width alone. (3)

Question 5 (8)

(a) Order: (1) deposit dielectric (ILD/low-k, CVD) → (2) pattern & etch trench (litho + dry etch) → (3) deposit barrier + Cu seed (PVD/ALD) → (4) electroplate Cu overfill → (5) CMP to planarize/remove overburden leaving Cu in trench. (4)

(b) Barrier (e.g. TaN/Ta by PVD or ALD) prevents Cu from diffusing into and poisoning the surrounding dielectric/Si; seed (thin Cu, by PVD/sputtering) provides a conductive, continuous surface for uniform electroplating to nucleate on. (4)

[
  {"claim":"DUV single-exposure half-pitch = 42.9 nm","code":"cd=0.30*193/1.35; result = abs(cd-42.888)<0.05"},
  {"claim":"EUV single-exposure half-pitch = 12.3 nm","code":"cd=0.30*13.5/0.33; result = abs(cd-12.27)<0.05"},
  {"claim":"Oxide net rise above original Si = 2.24 nm","code":"rise=4-0.44*4; result = abs(rise-2.24)<0.001"},
  {"claim":"Poisson yield = 0.9085 for D0=0.15, A=0.64","code":"Y=exp(-Rational(15,100)*Rational(64,100)); result = abs(float(Y)-0.90848)<0.001"},
  {"claim":"Dies per wafer approx 1021","code":"N=pi*150**2/64 - pi*300/sqrt(128); result = abs(float(N)-1021)<1.5"},
  {"claim":"Weighted die price = 154.25","code":"p=0.30*210+0.45*150+0.25*95; result = abs(p-154.25)<1e-9"},
  {"claim":"GAA stack Weff = 210 nm","code":"w=3*2*(30+5); result = w==210"}
]