TiN Thermo-Optic Phase Shifters

• Titanium nitride thermo-optic phase shifters are resistive microheaters integrated with SiN waveguides that modulate optical phase via electrically driven heating.
• They offer versatile implementation options including non-suspended and suspended architectures, balancing thermal coupling, optical isolation, and CMOS compatibility.
• Key performance metrics such as π-shift power, modulation bandwidth, and thermal trimming stability are optimized for reconfigurable photonic integrated circuits.

Searching arXiv for the cited papers to ground the article in the specified sources. Titanium nitride thermo-optic phase shifters are resistive microheaters integrated with silicon nitride photonic waveguides to control the phase of an optical signal through electrically driven heating and the resulting change in effective index. In the reported SiN platforms, TiN is used either in non-suspended heaters fabricated by a single-step lithographic metal definition process with aluminum interconnects, or in suspended heater bridges that simultaneously support low-power phase shifting and post-fabrication refractive-index trimming. The cited implementations span visible and near-infrared operation, including Mach–Zehnder interferometer platforms at $\lambda = 810 \,\mathrm{nm}$, $1550 \,\mathrm{nm}$, $561 \,\mathrm{nm}$, as well as related visible-light operation at $445 \,\mathrm{nm}$ and $532 \,\mathrm{nm}$ (Limongi et al., 9 Jun 2026, Chen et al., 30 Apr 2025).

1. Platform configurations and device geometries

The reported TiN thermo-optic phase shifters are implemented on SiN photonic integrated circuits using two distinct architectural classes. One class uses non-suspended heaters on top of oxide-clad SiN waveguides in two Mach–Zehnder-interferometer platforms targeting $810 \,\mathrm{nm}$ and $1550 \,\mathrm{nm}$. The other uses suspended bridges in a visible-light SiN platform, where the heater strip runs longitudinally above the waveguide and is laterally isolated by deep trenches (Limongi et al., 9 Jun 2026, Chen et al., 30 Apr 2025).

In the non-suspended devices, the substrate stack proceeds from a Si substrate to a bottom oxide, an LPCVD SiN waveguide core, a top oxide separating heater from waveguide, a PECVD overglass, and deep air trenches etched around the heater for lateral thermal isolation. Design 1 uses a $2.0 \,\mu\mathrm{m}$ bottom oxide, a $140 \,\mathrm{nm} \times 650 \,\mathrm{nm}$ SiN core, and a $1.8 \,\mu\mathrm{m}$ top oxide. Design 2 uses a $1550 \,\mathrm{nm}$0 bottom oxide, a $1550 \,\mathrm{nm}$1 SiN core, and a $1550 \,\mathrm{nm}$2 top oxide. In both cases, the TiN heater is centered over one arm of the MZI and consists of $1550 \,\mathrm{nm}$3 TiN sandwiched between two $1550 \,\mathrm{nm}$4 Ti adhesion layers; the heater width is $1550 \,\mathrm{nm}$5 for Design 1 and $1550 \,\mathrm{nm}$6 for Design 2 (Limongi et al., 9 Jun 2026).

In the suspended visible-light platform, the heater is formed on a bridge of total suspended length $1550 \,\mathrm{nm}$7 and width $1550 \,\mathrm{nm}$8, with each surrounding deep trench $1550 \,\mathrm{nm}$9 wide. The SiN waveguide beneath the TiN strip is $561 \,\mathrm{nm}$0 wide for single-mode operation at $561 \,\mathrm{nm}$1. The TiN film thickness is $561 \,\mathrm{nm}$2 with sheet resistance $561 \,\mathrm{nm}$3. The waveguide stack uses PECVD SiO$561 \,\mathrm{nm}$4 bottom and top claddings of thickness $561 \,\mathrm{nm}$5, a moderate-confinement SiN core of thickness $561 \,\mathrm{nm}$6, and a low-confinement SiN layer of thickness $561 \,\mathrm{nm}$7 for edge couplers (Chen et al., 30 Apr 2025).

Platform	Core structural features	TiN heater configuration
Non-suspended SiN MZI, 810 nm	$561 \,\mathrm{nm}$8 SiN, $561 \,\mathrm{nm}$9 top oxide, deep air trenches	$445 \,\mathrm{nm}$0 TiN between two $445 \,\mathrm{nm}$1 Ti layers, $445 \,\mathrm{nm}$2 width
Non-suspended SiN MZI, 1550 nm	$445 \,\mathrm{nm}$3 SiN, $445 \,\mathrm{nm}$4 top oxide, deep air trenches	$445 \,\mathrm{nm}$5 TiN between two $445 \,\mathrm{nm}$6 Ti layers, $445 \,\mathrm{nm}$7 width
Suspended visible-light SiN	$445 \,\mathrm{nm}$8 bridge, $445 \,\mathrm{nm}$9 width, $532 \,\mathrm{nm}$0 waveguide, deep trenches	$532 \,\mathrm{nm}$1 TiN strip along bridge

These geometries establish the central trade-off visible across the reported devices: heater-to-waveguide thermal coupling improves as the thermal path is reduced or isolated, while optical isolation and substrate leakage constraints can require thicker oxides or different cladding layouts. This suggests that TiN heater performance is inseparable from the full photonic and thermal stack rather than being determined by heater material alone.

2. Fabrication schemes and CMOS-compatible process integration

A principal result of the non-suspended implementation is a single-step lithographic process that defines both high-resistance TiN heaters and low-resistance Al interconnects within the same metal-patterning sequence. The process begins with bottom SiO$532 \,\mathrm{nm}$2 growth by thermal oxidation, followed by LPCVD SiN deposition and waveguide etching by i-line lithography and dry RIE. A top BPSG layer of $532 \,\mathrm{nm}$3 or $532 \,\mathrm{nm}$4 is then deposited and reflowed to planarize. Next, a multilayer stack of Al / $532 \,\mathrm{nm}$5 Ti / $532 \,\mathrm{nm}$6 TiN / $532 \,\mathrm{nm}$7 Ti is sputter-deposited. A single photolithographic step defines narrow heater lines and wide Al interconnects or pads, after which the full metal stack is dry-etched anisotropically down to the BPSG. A selective wet etch using PAN removes Al laterally from features narrower than $532 \,\mathrm{nm}$8, leaving narrow features as Ti–TiN–Ti only and wide features as Al–Ti–TiN–Ti. PECVD SiO$532 \,\mathrm{nm}$9 overglass is deposited, bond-pad windows are opened by lithography and dry etch, and chip facets and deep trenches are defined by Bosch etching through Si. The reported process uses standard CMOS-compatible tools—LPCVD, PVD, PECVD, i-line lithography, and RIE—and avoids multi-step metal and post-release MEMS processing (Limongi et al., 9 Jun 2026).

The suspended visible-light devices follow a different flow because the heater must be thermally isolated from the substrate. On 200 mm Si wafers, the process includes Si mesa patterning and doping for on-chip photodiodes, PECVD SiO$810 \,\mathrm{nm}$0/SiN stack deposition, photolithographic definition of the two SiN waveguide layers, CMP planarization, TiN sputter deposition and lift-off for heater definition, two levels of Al metallization and vias for electrical routing, and a deep-trench etch plus isotropic Si undercut to form suspended bridges and facets for edge coupling (Chen et al., 30 Apr 2025).

The contrast between these flows is technically consequential. The non-suspended process prioritizes lithographic simplification and foundry compatibility, whereas the suspended process prioritizes thermal isolation through release and undercut. A common misconception is that TiN-based thermo-optic phase shifters imply a single fabrication paradigm; the cited devices show instead that TiN is compatible with materially different integration strategies, from fully oxide-embedded heaters to released bridges.

3. Governing relations for phase tuning and thermal dynamics

The non-suspended MZI phase shifters are described by the thermo-optic relation

$810 \,\mathrm{nm}$1

with

$810 \,\mathrm{nm}$2

so that $810 \,\mathrm{nm}$3 for small $810 \,\mathrm{nm}$4. Their thermal tuning efficiency is defined as

$810 \,\mathrm{nm}$5

in units of $810 \,\mathrm{nm}$6. For the MZI transfer function, the reported dependence is

$810 \,\mathrm{nm}$7

which can be inverted to retrieve $810 \,\mathrm{nm}$8. Thermal dynamics are approximated as a single-pole RC system with thermal time constant $810 \,\mathrm{nm}$9 and cut-off

$1550 \,\mathrm{nm}$0

Good agreement between $1550 \,\mathrm{nm}$1 and $1550 \,\mathrm{nm}$2 was reported for both wavelength platforms, confirming single-pole behavior (Limongi et al., 9 Jun 2026).

The suspended platform is analyzed using both heat-transport and effective-index relations. In steady state, the heater cross-section satisfies Fourier’s law,

$1550 \,\mathrm{nm}$3

where $1550 \,\mathrm{nm}$4 is the thermal conductivity of SiN, SiO$1550 \,\mathrm{nm}$5, TiN, and Si, and $1550 \,\mathrm{nm}$6 is the volumetric Joule heating in the TiN. A one-dimensional thermal-resistance approximation gives

$1550 \,\mathrm{nm}$7

The thermal-isolation factor is defined as

$1550 \,\mathrm{nm}$8

From COMSOL simulation, $1550 \,\mathrm{nm}$9, while the corresponding non-suspended geometry would require $2.0 \,\mu\mathrm{m}$0 to reach the same $2.0 \,\mu\mathrm{m}$1, giving $2.0 \,\mu\mathrm{m}$2 (Chen et al., 30 Apr 2025).

The same suspended study reports the phase-shift response at constant wavelength as

$2.0 \,\mu\mathrm{m}$3

along with the temperature-dependent effective-index relation

$2.0 \,\mu\mathrm{m}$4

For the core, published values for stoichiometric Si$2.0 \,\mu\mathrm{m}$5N$2.0 \,\mu\mathrm{m}$6 give $2.0 \,\mu\mathrm{m}$7 at visible wavelengths, and the effective-index sensitivity is $2.0 \,\mu\mathrm{m}$8 (Chen et al., 30 Apr 2025).

Taken together, these relations show that TiN phase shifters are governed by coupled electrical, thermal, and modal physics. This suggests that comparative figures such as $2.0 \,\mu\mathrm{m}$9 and bandwidth should be interpreted relative to thermal boundary conditions and optical stack design, not as material constants of TiN.

4. Electro-optical performance in non-suspended SiN platforms

The single-step non-suspended devices were characterized electro-optically on MZI platforms at $140 \,\mathrm{nm} \times 650 \,\mathrm{nm}$0 and $140 \,\mathrm{nm} \times 650 \,\mathrm{nm}$1. Safe operating area was determined from IV sweeps up to failure. For Design 1, the reported maxima are $140 \,\mathrm{nm} \times 650 \,\mathrm{nm}$2, $140 \,\mathrm{nm} \times 650 \,\mathrm{nm}$3, and $140 \,\mathrm{nm} \times 650 \,\mathrm{nm}$4. For Design 2, the values are $140 \,\mathrm{nm} \times 650 \,\mathrm{nm}$5, $140 \,\mathrm{nm} \times 650 \,\mathrm{nm}$6, and $140 \,\mathrm{nm} \times 650 \,\mathrm{nm}$7 (Limongi et al., 9 Jun 2026).

The $140 \,\mathrm{nm} \times 650 \,\mathrm{nm}$8-shift powers were extracted from DC voltage-ramp measurements combined with MZI inversion. The reported values are $140 \,\mathrm{nm} \times 650 \,\mathrm{nm}$9 for Design 1 at $1.8 \,\mu\mathrm{m}$0 and $1.8 \,\mu\mathrm{m}$1 for Design 2 at $1.8 \,\mu\mathrm{m}$2. Small-signal sine-wave measurements from $1.8 \,\mu\mathrm{m}$3 to $1.8 \,\mu\mathrm{m}$4 yielded modulation bandwidths of $1.8 \,\mu\mathrm{m}$5 for Design 1 and $1.8 \,\mu\mathrm{m}$6 for Design 2. Time-domain fits to

$1.8 \,\mu\mathrm{m}$7

under $1.8 \,\mu\mathrm{m}$8 square-wave drive gave $1.8 \,\mu\mathrm{m}$9 and $1550 \,\mathrm{nm}$00 for Design 1, and $1550 \,\mathrm{nm}$01 and $1550 \,\mathrm{nm}$02 for Design 2 (Limongi et al., 9 Jun 2026).

The reported interpretation links these differences directly to the oxide thicknesses and thermal loading. The thinner top oxide of $1550 \,\mathrm{nm}$03 in Design 1 yields stronger thermal coupling, lower $1550 \,\mathrm{nm}$04, and faster $1550 \,\mathrm{nm}$05. The thicker oxide in Design 2 is used to trade optical isolation—specifically $1550 \,\mathrm{nm}$06 substrate leakage—against increased thermal inertia, which raises $1550 \,\mathrm{nm}$07 and slows $1550 \,\mathrm{nm}$08. Deep air trenches improve lateral thermal confinement and reduce crosstalk (Limongi et al., 9 Jun 2026).

A recurrent misunderstanding is to equate lower $1550 \,\mathrm{nm}$09 with universally better device behavior. In the reported data, lower $1550 \,\mathrm{nm}$10 is accompanied by design choices that can alter optical isolation, substrate leakage, and thermal confinement; the performance space is therefore explicitly multi-objective rather than one-dimensional.

5. Suspended TiN heaters as low-power phase shifters and index trimmers

The suspended TiN heaters support a dual operating mode. As phase shifters, they provide compact visible-light tuning with low power dissipation. As overdriven heaters, they induce persistent refractive-index trimming in the surrounding photonic structure. In the reported AMZI at $1550 \,\mathrm{nm}$11, a single-pass $1550 \,\mathrm{nm}$12 phase shifter yields $1550 \,\mathrm{nm}$13, with insertion loss $1550 \,\mathrm{nm}$14 for the phase-shifter section, AMZI insertion loss $1550 \,\mathrm{nm}$15, and extinction $1550 \,\mathrm{nm}$16. In a three-pass suspended-bridge MZI switch at $1550 \,\mathrm{nm}$17, $1550 \,\mathrm{nm}$18. Related earlier work cited in the source reports $1550 \,\mathrm{nm}$19 at $1550 \,\mathrm{nm}$20 and $1550 \,\mathrm{nm}$21 (Chen et al., 30 Apr 2025).

The same structures are driven at $1550 \,\mathrm{nm}$22 to $1550 \,\mathrm{nm}$23 for thermal trimming. COMSOL indicates that the SiN waveguide beneath the heater reaches $1550 \,\mathrm{nm}$24 at $1550 \,\mathrm{nm}$25, while the TiN itself is only $1550 \,\mathrm{nm}$26 hotter. Outside the $1550 \,\mathrm{nm}$27 bridge, temperature falls by hundreds of degrees over $1550 \,\mathrm{nm}$28, enabling localized trimming. At $1550 \,\mathrm{nm}$29, the observed modal effective-index changes are $1550 \,\mathrm{nm}$30 after $1550 \,\mathrm{nm}$31 at $1550 \,\mathrm{nm}$32 with simulated $1550 \,\mathrm{nm}$33, and $1550 \,\mathrm{nm}$34 after $1550 \,\mathrm{nm}$35 at $1550 \,\mathrm{nm}$36. These changes were stable over an observation period of 97 days, with phase relaxation after 97–118 days storage of $1550 \,\mathrm{nm}$37, against a reported measurement error of $1550 \,\mathrm{nm}$38 (Chen et al., 30 Apr 2025).

The trimming procedure is explicitly operationalized: moderate over-drive power of $1550 \,\mathrm{nm}$39 to $1550 \,\mathrm{nm}$40 is applied for tens of minutes, heater power is periodically removed, the device is allowed to cool for approximately 1–2 minutes, and transmission is measured using wafer-scale laser or supercontinuum plus tunable filter while fringe shift $1550 \,\mathrm{nm}$41 or port imbalance is tracked in a $1550 \,\mathrm{nm}$42 MZI. Using $1550 \,\mathrm{nm}$43 for 74–82 minutes, the bias power of five thermo-optic MZI devices at $1550 \,\mathrm{nm}$44 was reduced from $1550 \,\mathrm{nm}$45 to $1550 \,\mathrm{nm}$46 (Chen et al., 30 Apr 2025).

This dual-mode operation distinguishes the suspended architecture from the non-suspended platform. A plausible implication is that TiN heaters can function not only as volatile tuning elements but also as localized post-fabrication correction elements when the surrounding stack supports a stable thermally induced material modification.

6. Mechanistic interpretation, trade-offs, and circuit relevance

The reported mechanistic interpretation of persistent trimming centers on the cladding rather than the SiN core. TEM/EDX confirms a SiN core with negligible O and a SiO$1550 \,\mathrm{nm}$47 cladding with negligible N. Ellipsometry on blanket $1550 \,\mathrm{nm}$48 SiN/$1550 \,\mathrm{nm}$49 SiO$1550 \,\mathrm{nm}$50 gives $1550 \,\mathrm{nm}$51 at $1550 \,\mathrm{nm}$52, consistent with nitrogen-rich SiN. FTIR before and after annealing up to $1550 \,\mathrm{nm}$53 shows negligible change in SiN-bond and SiH and NH peaks at $1550 \,\mathrm{nm}$54, while the SiO peak near $1550 \,\mathrm{nm}$55 blueshifts and grows in intensity at $1550 \,\mathrm{nm}$56, and the SiOH peak near $1550 \,\mathrm{nm}$57 decreases markedly at $1550 \,\mathrm{nm}$58. The stated hypothesis is that heat-driven dehydroxylation of the PECVD SiO$1550 \,\mathrm{nm}$59 cladding, $1550 \,\mathrm{nm}$60, creates nano-porosity and lowers $1550 \,\mathrm{nm}$61, thereby lowering the guided-mode $1550 \,\mathrm{nm}$62 by up to $1550 \,\mathrm{nm}$63. Core-only mechanisms such as Si–H or N–H bond breaking are reported to appear negligible below $1550 \,\mathrm{nm}$64 for nitrogen-rich SiN (Chen et al., 30 Apr 2025).

The suspended-heater study also introduces a kinetics model,

$1550 \,\mathrm{nm}$65

$1550 \,\mathrm{nm}$66

where $1550 \,\mathrm{nm}$67 is a Gaussian defect-energy distribution with mean $1550 \,\mathrm{nm}$68, FWHM $1550 \,\mathrm{nm}$69, and $1550 \,\mathrm{nm}$70. From this demarcation-energy fit, the predicted $1550 \,\mathrm{nm}$71 over one year at $1550 \,\mathrm{nm}$72 is $1550 \,\mathrm{nm}$73 to $1550 \,\mathrm{nm}$74, and over five years is $1550 \,\mathrm{nm}$75 to $1550 \,\mathrm{nm}$76 (Chen et al., 30 Apr 2025).

For the non-suspended platform, the reported circuit implication is different: the single-step metal definition eliminates extra masks and post-processing, thereby easing integration into CMOS foundry flows. The demonstrated figures—$1550 \,\mathrm{nm}$77 and $1550 \,\mathrm{nm}$78 in the kHz range—are described as competitive for non-suspended SiN heaters and sufficient for reconfigurable photonic processors, microwave photonics, and quantum circuits where kHz-range tuning is acceptable. Future optimizations such as substrate undercut could further lower $1550 \,\mathrm{nm}$79, at the expense of process complexity or dynamic speed (Limongi et al., 9 Jun 2026).

These results frame the main design controversy in practical terms rather than as a dispute over heater material. Suspended TiN heaters achieve very low $1550 \,\mathrm{nm}$80 through strong thermal isolation and can enable trimming, but they require release processing. Non-suspended TiN heaters preserve a simpler CMOS-compatible flow and avoid post-release MEMS processing, but operate at substantially higher $1550 \,\mathrm{nm}$81. The reported literature therefore supports no universal ranking of TiN heater architectures independent of application constraints.

7. Position within reconfigurable SiN photonics

Across the cited implementations, TiN thermo-optic phase shifters serve as building blocks for reconfigurable SiN photonic integrated circuits, especially MZI-based processors and switches. In the non-suspended case, they are integrated with aluminum interconnects and bond pads through a single lithographic metal-definition step, directly targeting scalable foundry-compatible fabrication (Limongi et al., 9 Jun 2026). In the suspended visible-light case, they support both dynamic phase control and permanent bias correction, reducing the bias power required to null one output port of thermo-optic MZI devices from $1550 \,\mathrm{nm}$82 to $1550 \,\mathrm{nm}$83 after trimming (Chen et al., 30 Apr 2025).

The body of results indicates that TiN is effective across disparate SiN operating regimes: non-suspended near-infrared and visible/near-visible platforms with oxide isolation, and suspended visible-light bridges with aggressive thermal isolation. This suggests that the principal research axis is not whether TiN can function as a thermo-optic heater, but how lithographic simplification, oxide thickness, trenching, suspension, and thermal confinement should be co-optimized for a given wavelength range, bias budget, dynamic requirement, and integration flow.