5.7 Patterning: lithography, masks, and etching
L1 Phenomenon
LN is hard to etch: fluorine chemistries form non-volatile LiF, so most low-loss TFLN waveguides are cut by argon ions that knock atoms off physically. Physical etching removes the mask too, and some of the removed LN lands back on the sidewalls. Mask choice, ion energy, and chamber pressure decide how steep and smooth the sidewalls are, and the sidewalls decide the loss.
xxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxx
xxxxx................xxxxx
xxxx......................xxxx
xxxx..........................xxxx
xxxx............................xxxx
xxxx..########+########+########..xxxx
xxx...########+########+########...xxx
xxx....########+########+########....xxx
xxx....++++++++++++++++++++++++++....xxx
xxx....++++++++++++++++++++++++++....xxx
xxx....########+########+########....xxx
xxx...########+########+########...xxx
xxxx..########+########+########..xxxx
xxxx............................xxxx
xxxx..........................xxxx
xxxx......................xxxx
xxxxx................xxxxx
xxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxx - E-beam, whole wafer
- 4.9 d
- Stepper, whole wafer
- under 1 min
Usable area: wafer minus 8 mm edge exclusion; exposed area 10% of it (template). Step size, settling and stage moves are not included, so e-beam times are lower bounds. 22 × 22 mm fields: 6 per wafer derived. Stepper fields fill uniformly over 60 s, the reported upper bound (order template). E-beam path drawn at 4 mm pitch; its length fraction equals the written fraction derived.
dddddddddddddd
dddddddddddddd
dddddddddddddd
dddddddddddddd
LLLLLLLLLLLLLL
~LLLLLLLLLLLLLL~
~LLLLLLLLLLLLLL~
~LLLLLLLLLLLLLL~
~LLLLLLLLLLLLLL~
~LLLLLLLLLLLLLL~
oooooooooooooooooooooooooooooooooooooooooooo
oooooooooooooooooooooooooooooooooooooooooooo Ar+ flux from the top; mask dark, LN straw, redeposition grey (template thickness); not to scale.
- Model sidewall
- 88.3°
- Nearest measured
- DLC mask, 80°
Green curve: model upper bound for the 85.0° mask; dots: measured sidewalls P P.
| # | STEP | ROUTE |
|---|---|---|
| 13 | PECVD SiO2 hard mask | ridge |
| 14 | Spin ARC and DUV photoresist | ridge |
| 15 | DUV stepper exposure, step-and-repeat | ridge |
| 16 | Develop; etch ARC (Ar/O2) and SiO2 (C3F8) | ridge |
| 17 | Ar ICP-RIE of LN (ridge) | ridge |
| 18 | Strip resist and SiO2 hard mask in HF | ridge |
| 13 | PECVD DLC hard mask (from CH4) with Si3N4 intermediate mask | strip |
| 14 | DUV stepper lithography | strip |
| 15 | Transfer into Si3N4 (fluorine plasma), then into DLC (oxygen plasma) | strip |
| 16 | Ar ion beam etching, full etch to the BOX | strip |
| 17 | SC-1 redeposition removal; O2 plasma DLC strip | strip |
rrrrrr
rrrrrr
rrrrrr
aaaaaa
oooooo
oooooo
oooooo
oooooo
~LLLLLL~
~LLLLLLLL~
LLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLL
LLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLL
oooooooooooooooooooooooooooooooooooooooooooooooo
oooooooooooooooooooooooooooooooooooooooooooooooo
oooooooooooooooooooooooooooooooooooooooooooooooo
SSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSS
SSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSS
SSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSS
SSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSS17/23Ar ICP-RIE of LN (ridge) (ridge route)
L2 Model and data
Lithography
| Item | Value | Source; status |
|---|---|---|
| EBL, HSQ (FOx-16) negative resist, 4-inch LNOI, 100 kV | about 500 nm HSQ; sleeve 150 nm wide, 4 passes at 1900 µC/cm2, 8 nA, 5 nm step; bulk 7600 µC/cm2, 100 nA, 25 nm step; develop 5 min in 1:3 AZ351B:DI water; resist sidewall 85° | P |
| EBL, ZEP-520A positive resist | about 1 µm, baked 170 °C for 2 min; 300 µC/cm2 at 2 nA (JEOL 9500) | P |
| EBL, HSQ multi-pass | HSQ with multi-pass exposure | P |
| DUV stepper | 248 nm; 22 mm × 22 mm reticle stepped across 4- and 6-inch wafers; 60 nm ARC + 600 nm resist on 650 nm PECVD SiO2 hard mask | P |
| DUV vs EBL write time | DUV exposure under 1 min per wafer vs more than 8 days for EBL at 10 nA, 125 kV (authors' estimate) | P |
| EBL write time, bulk dose above, per cm2 | 21.1 h at 100 nA | derived t = D A / I = 7600 µC/cm2 × 1 cm2 / 100 nA |
| Stepper for poling electrodes | 0.15 µm and 0.5 µm steppers for combs and pads | P |
| Femtosecond-laser direct writing (PLACE) | ablates a 600 nm sputtered Cr film into a hard mask | P |
Hard masks
| Mask | Reported use | Source; status |
|---|---|---|
| HSQ (spin-on oxide) | direct EBL resist and mask for Ar+ etching; microdisks with HSQ masks from 2014 P(title) | P, P, P |
| SiO2 (PECVD, 650 nm) | DUV pattern transferred by C3F8 etch; removed with HF | P |
| a-Si (deposited Si) | EBL then Ar+ plasma etch of LN; also used for nanophotonic SHG waveguides P(title) | P; outside5sigma.com note: a-Si mask cracked and etched inhomogeneously on Y-cut reported |
| Cr (lift-off) | EBL lift-off Cr mask; F + Ar plasma; about 75° sidewalls, < 2 nm RMS roughness, 0.4 dB/cm | P |
| Cr (sputtered, 600 nm, fs-laser patterned) | mask for CMP-based pattern transfer (PLACE) | P |
| DLC (PECVD from CH4, 300 nm) | DUV pattern into Si3N4 (F chemistry), then DLC (O2 plasma); LN:DLC selectivity up to 3 vs about 1 for SiO2 masks; 80° sidewalls | P |
| ZEP (polymer) | direct mask for Ar ion milling, slanted sidewalls | P |
Argon etching: reported conditions
| Tool and conditions | Result | Source; status |
|---|---|---|
| ICP, Ar 10 sccm, 1 mTorr, 20 °C, DC bias swept 100-800 V | redeposition falls and trenching rises with bias | P |
| ICP 300 W, 1 mTorr, 600 V | 24 nm/min | P |
| Pressure sweep 1-15 mTorr (400 V, 600 W) and 0.3-13 mTorr (600 V, 300 W) | etch rate and sidewall angle fall with pressure; lateral etching dominates near 9 mTorr (400 V) and 7 mTorr (600 V); redeposition-free near 11 mTorr (400 V) and 5-7 mTorr (600 V); minimum rates 27 and 12 nm/min | P |
| ICP-RIE Ar+, bias power 112 W | 350 nm etched, 250 nm slab; Qi 10.0 ± 0.7 million | P |
| ICP-RIE Ar plasma | about 80 nm/min; 200 nm of a 400 nm film | P |
| Ion mill, 600 V beam | about 35 nm/min; smooth, slanted sidewalls | P |
| Ar IBE with DLC mask | full etch of 0.7 µm, base width 3.3 µm, 80° | P |
| IBE, incidence angles 0, 15, 25, 35° | used to test the redeposition argument | P |
Redeposition removal
| Method | Value | Source; status |
|---|---|---|
| SC-1 wet clean | NH4OH:H2O2:H2O = 1:1:5 | P |
| KOH:H2O2 (LiTaO3 analog) | 3:1 KOH(40%):H2O2(30%) | P |
| Dry-only regimes | three methods: tuned bias and pressure; dense patterns or a retained slab; trapezoidal mask | P |
Sidewall angles reported
| Angle | Context | Source; status |
|---|---|---|
| about 60° | stated typical for ridge waveguides with SiO2 masks | P |
| about 75° | Cr mask, F + Ar | P |
| 80° | DLC mask, Ar IBE | P |
| 85° | HSQ resist after development (before etch) | P |
Alternatives to Ar etching
| Method | Reported | Source; status |
|---|---|---|
| Fluorine RIE | LiF by-products are non-volatile and impede etching | P, P |
| PLACE (photolithography-assisted chemo-mechanical etching) | 700 nm z-cut; Cr mask by fs laser; CMP transfers the pattern; Cr wet strip; second CMP; anneal 450 °C, 2 h, air; ring R = 200 µm, gap 3.8 µm; QL 4.29 × 106, Qi 4.04 × 107, < 1 dB/m | P |
| PLACE microdisk | Q > 108; OPO threshold 19.6 µW | P(abs) |
| PLACE origin (microdisks) | Q above 107 | P(title) |
Loss and Q by fabrication route
| Route | Best value found | Source; status |
|---|---|---|
| EBL + Ar ICP-RIE, ion-sliced, ring | Qi 2.9 × 107, 1.3 dB/m | P(abs) |
| EBL + Ar ICP-RIE, ring (2017 baseline) | Qi about 1 × 107, 2.7 dB/m | P |
| DUV + SiO2 mask + Ar ICP-RIE, wafer-scale | 0.27 ± 0.05 dB/cm across a 4-inch wafer | P |
| DUV + DLC + Ar IBE, strip | down to 4 dB/m | P |
| HSQ + tuned Ar ICP, redeposition-free | 1.55 dB/cm | P |
| PLACE ring with bus | < 1 dB/m (Qi 4.04 × 107) | P(abs) |
| CMP film from bulk, ring | Q about 1 × 108 | P |
| Material limit after anneal | about 0.2 dB/m (material-limited Q 180 million in arXiv v1) | P |
| Zhang 2017-12, EBL + Ar | 2.70 dB/m | P |
| Luke 2020-08, DUV + Ar | 27 dB/m | P |
| Gao 2022, CMP from bulk | 0.40 dB/m | template |
| Li 2023-08, DUV + DLC | 4.00 dB/m | P |
| Kaufmann 2023-01, EBL + Ar | 155 dB/m | P |
| Li 2023-09, PLACE | <1.00 dB/m | P(abs) |
| Zhu 2024-07, EBL + Ar | 1.30 dB/m | P(abs) |
Green staircase: lowest reported loss up to each date among the rows shown (Q-only rows excluded). Q axis (log only): Q = 2π ng / (λ α), λ = 1550 nm, ng = 2.3 template. Dashed arrow: anneal of ion-sliced LN from about 1.5 to 0.2 dB/m P. Arrow under a marker: upper bound. Hover or focus a marker for its paper.
L3 Details and open points
- Etched vs etchless: bonding TFLN onto patterned Si or SiN avoids LN etching altogether; the trade-offs are reviewed in P(title).
- Why the angle matters: a slanted ridge needs a wider base, raises bend radius, and reduces electrode overlap; vertical strips allow tighter integration P.
- Sidewall model used in the etch figure (derived geometry, inputs template unless a row above supplies them): a mask edge of slope angle am receding vertically at rate Rm retreats laterally at Rm / tan(am); the LN etched at RLN then takes slope tan(aLN) = S tan(am) with S = RLN / Rm. With S = 1 and am = 85° the model gives 85°; measured sidewalls are lower because sputter yield depends on incidence angle and redeposition builds a second slope P. The model is an upper bound.
- Etch rates are tool-specific; Kaufmann et al. vary ICP power only to keep the endpoint system in range and report that power has only a minor effect on redeposition at fixed bias and pressure P.
- Charging: a 5-10 nm Al discharge layer on dielectrics such as LN during EBL is reported in the author's blog (outside5sigma.com) reported.
- Pyroelectric hazard: heating z-cut LN on a hotplate can expose resist by pyroelectric discharge and crack substrates; outside5sigma.com notes reported, reported. Pyroelectric coefficient magnitude 4 × 10−5 C/(m2 K) in one compilation P-comp and 8.3 × 10−5 C/(m2 K) from a vendor V.