MENU
Advanced Materials Intelligence
  1. Home
  2. Optical Modulation and Modulation Formats Explained | Photonics

Optical Modulation and Modulation Formats Explained | Photonics

Back to Technology, Institution & Company Guides

TECHNOLOGY EXPLAINER

Optical Modulation and Modulation Formats
— putting information on light is a materials problem: how far can you move the refractive index?

Optical communication puts information onto laser light by working on its brightness, phase and polarisation. NRZ carries one bit per symbol, PAM4 two, and DP-16QAM, which also uses polarisation, eight. The part that does the work is the modulator, and there are three main types: Mach-Zehnder, electro-absorption and microring. Every one of them comes down to using a voltage to change a material's refractive index or absorption very slightly. Which material, and by how much — that is where the contest is.

Built from primary sources: public documents from IEEE 802.3, OIF, the LPO MSA and the OSFP MSA, and peer-reviewed papers in Nature and Physical Review Letters / Last updated September 2026

Conceptual image of a single red beam of light passing through a long, thin, clear crystal plate against a dark background, flickering slightly between light and dark as it leaves
Conceptual image (AI-generated). An impression of light passing through a material being worked on electrically. It does not show the structure, material, dimensions or light colour of a real modulator.
What this article covers
  1. What optical modulation is (the short version)
  2. Modulation formats — NRZ, PAM4 and QAM differ in how many bits each symbol carries
  3. Our calculation: the price of PAM4, and the bit arithmetic of QAM
  4. How modulators work (1): Mach-Zehnder — turning phase into light and dark by interference
  5. How modulators work (2): electro-absorption and microrings
  6. A materials engineer's view (1): the refractive index barely moves
  7. A materials engineer's view (2): the material decides how the effect works
  8. What is still hard
  9. Glossary / References / Claim-to-source audit
How claims are labelled in this article

Sourced = stated in a standard, a public document from a standards body, or a peer-reviewed paper (link given)
Our calculation = a figure this article derived, with the assumptions spelled out
Not yet confirmed = still in development or at the research stage, with nothing settled and no confirmed volume production
General explanations of the physics and materials-design interpretations are marked separately as Commentary.

1. What optical modulation is (the short version)

  • What it does: it takes laser light shining at a constant level and varies its brightness (intensity), the offset of its peaks and troughs (phase) and the direction in which it oscillates (polarisation) in step with an electrical signal, so that the light carries information (commentary)
  • How the information goes on (the modulation format): short data-centre links use NRZ and PAM4, which encode data in brightness levels; long-haul links use QAM, which also uses phase and polarisation. The LPO specification for 100 Gb/s per lane uses 53.125 GBd PAM4Sourced, and OIF's 400ZR uses DP-16QAMSourced
  • What puts it on (the modulator): either the laser's drive current is varied directly, or a separate modulator outside the laser works on the light. The main modulator types are Mach-Zehnder, electro-absorption and microring (commentary)
How this fits with the rest of the series

Our silicon photonics explainer mentioned that imec developed three kinds of silicon modulator: ring, Mach-Zehnder and GeSi electro-absorption. This article looks, from the materials side, at why and how those three change light. How QAM is received and recovered is covered in our explainer on coherent optical communication and DSP, thin-film lithium niobate in our thin-film lithium niobate explainer, integrating InP modulators with lasers in our explainer on photonic integrated circuits and indium phosphide, and the use of microrings for wavelength multiplexing in our explainer on wavelength-division multiplexing and microrings.

2. Modulation formats — NRZ, PAM4 and QAM differ in how many bits each symbol carries

An optical signal is a sequence of symbols, each occupying a fixed slot of time. The number of symbols per second is the baud rate (Bd, GBd), and what distinguishes one modulation format from another is how many bits each symbol carries (commentary).

How much information one symbol carries (schematic) Left two: brightness levels / right: combinations of phase and amplitude (points on the complex plane) NRZ (2 levels) PAM4 (4 levels) 16QAM (16 points) 1 0 10 11 01 00 1 symbol = 1 bit 1 symbol = 2 bits 1 symbol = 4 bits light on or off four brightness levels 8 bits with two polarisations (DP) Note: the 2-bit mapping for PAM4 (Gray code) is one example for illustration. Waveforms and point positions are schematic. Note: data-centre 100G and 200G lanes use PAM4; OIF 400ZR and 800ZR use DP-16QAM (see the sources in the text). Note: NRZ and PAM4 use brightness only and are paired with receivers that measure it directly (direct detection).
Fig. 1 Conceptual diagram (vector drawing). The use of PAM4 follows the LPO MSA specification [Ref. 3] and the OSFP specification [Ref. 4]; the use of DP-16QAM follows the OIF 400ZR [Ref. 5] and 800ZR [Ref. 6] implementation agreements. Waveforms, point positions and bit mappings are schematics of the general principle, not the detailed definitions in each standard.
Application / standardModulation formatSymbol rateWhat the document says
100G-DR-LPO (LPO MSA)PAM453.125 GBd100 Gb/s per lane, up to 500 m over SMF
1.6TAUI-8 (electrical lanes in the OSFP specification)224G-PAM4106.25 GBaud8 lanes × 200 Gb/s = 1.6 Tb/s
400ZR (OIF)DP-16QAM59.84375 GBaud per polarisation478.750 Gbps in total. For amplified links up to 120 km and other uses
800ZR (OIF)DP-16QAM118.203350603 GbaudSingle-span, amplified DWDM over 80 to 120 km

All Sourced (LPO MSA specification [Ref. 3], OSFP specification [Ref. 4], OIF 400ZR implementation agreement Rev 3.0 [Ref. 5], OIF 800ZR implementation agreement Rev 1.0 [Ref. 6]). 1.6TAUI-8 values are for electrical lanes.

3. Our calculation: the price of PAM4, and the bit arithmetic of QAM

Put more bits on each symbol and you can send more data at the same baud rate. But it does not come free. The calculation below shows what is being paid.

PAM4 halves the speed but cuts the gap to a third (our calculation) The gaps within the same amplitude 1 1/3 1/3 1/3 NRZ: one gap PAM4: three gaps 20 log₁₀(1/3) = −9.5 dB To build a 200 Gb/s-class lane PAM4 (2 bits/symbol) 106.25 GBd NRZ at the same bit rate 212.5 GBd DP-16QAM (8 bits/symbol) 400ZR: 59.84375 x 8 478.75 Gb/s 800ZR: 118.2 x 8 about 945.6 Gb/s All include error-correction and other overhead bits Note: 106.25 GBaud is from the OSFP spec; 59.84375 GBaud and 478.750 Gbps from OIF 400ZR; 118.203350603 Gbaud from OIF 800ZR. Note: −9.5 dB, 212.5 GBd and 945.6 Gb/s are our calculations; −9.5 dB assumes four ideal, evenly spaced levels. Note: real signal quality also depends on noise, nonlinearity and bandwidth limits, not on this ratio alone.
Fig. 2 Drawing that includes our calculation (vector drawing). The baud rates and 478.750 Gbps are stated in the OSFP specification [Ref. 4], OIF 400ZR [Ref. 5] and 800ZR [Ref. 6]. −9.5 dB, 212.5 GBd and about 945.6 Gb/s are this article's calculations, not published values.
Our calculation: what is paid, and what is gained
  • The price of PAM4: put four evenly spaced levels into the same brightness range and the gap between neighbours is 1/3 of NRZ's. In amplitude terms that is 20 log10(1/3) = about −9.5 dBOur calculation
  • The return on PAM4: 1.6TAUI-8 runs at 106.25 GBaud × 2 bits = 212.5 Gb/s. Sending the same bits with NRZ would take 212.5 GBdOur calculation
  • The bit arithmetic of QAM: 16QAM has 16 points, so log216 = 4 bits; using two polarisations gives 8 bits per symbol. For 400ZR, 59.84375 × 8 = 478.75 Gb/s, which matches OIF's stated 478.750 GbpsOur calculation. For 800ZR, 118.203 × 8 = about 945.6 Gb/sOur calculation

Assumptions and limits: −9.5 dB is a geometric ratio assuming four ideal, evenly spaced levels. The real performance difference with PAM4 depends on the kind of noise, modulator nonlinearity and receiver equalisation. The calculated bit rates include error correction (FEC) and framing overhead, so they are higher than the data rate available to users (400 Gb/s and so on).

So PAM4 is a format that, instead of asking the electronics and the modulator to go twice as fast, asks them to produce steps three times finer, accurately. That ability to produce steps accurately — modulator linearity — is set by materials and device structure (our commentary). The non-linear compensation in the LPO specification, discussed in our explainer on NPO and LPO, is precisely a mechanism for correcting distortion in those steps.

4. How modulators work (1): Mach-Zehnder — turning phase into light and dark by interference

A Mach-Zehnder (MZ) modulator splits light into two paths, shifts the phase in one path (or both) and then recombines them. If the phases match, the waves reinforce each other and the output is bright; if they are half a wavelength (π) apart, they cancel and the output is dark (commentary).

Mach-Zehnder modulator (schematic, top view) A voltage on the electrode changes the refractive index of the waveguide beneath it and shifts the phase of the light Electrodes (signal voltage) Phase accumulates along the arm length L The other arm (reference) Input light Output light split combine Phase difference 0: reinforce, bright the crests of the two waves line up Phase difference π: cancel, dark crest meets trough Note: schematic only. Electrode layout (single arm, both arms, travelling wave) and waveguide material and size vary by device.
Fig. 3 Conceptual diagram (vector drawing). A schematic of the general principle of intensity modulation by interference. The layout and dimensions of electrodes and waveguides do not represent any particular device.

The strength of the MZ type is that it can manipulate phase itself, not just brightness. Combine two MZs and you can place points freely in phase and amplitude, as in the QAM of Section 2 (commentary). Its weakness is length. Because the refractive index of a material changes only slightly, shifting the phase by π needs arms of a certain length (calculated in Section 6).

Which material does the phase shifting changes the character of the device a great deal.

  • Lithium niobate (LiNbO₃): a crystal with an electro-optic effect (the Pockels effect) in which the refractive index changes with voltage. In Nature (2018), Wang et al. describe conventional LiNbO₃ modulators as "bulky, expensive, limited in bandwidth and require high drive voltages", and with a thin-film, integrated modulator demonstrated CMOS-compatible drive voltages, data rates of up to 210 Gb/s and on-chip optical loss of less than 0.5 dBSourced
  • Silicon: in Nature (2004), Liu et al. noted that silicon waveguide modulators had until then reached modulation frequencies of only about 20 MHz, and demonstrated an all-silicon modulator with an embedded MOS capacitor structure and a modulation bandwidth exceeding 1 GHzSourced. The paper says the device is compatible with CMOS processing and opens the way to monolithic integration with advanced electronics on the same silicon substrateSourced

5. How modulators work (2): electro-absorption and microrings

How electro-absorption and microring modulators work (schematic) Electro-absorption (EA) Microring (MR) operating wavelength absorption wavelength (longer to the right) no voltage with voltage The field pushes the absorption edge longer; light is absorbed electrode ring input output The resonance dip (dark wavelength) shifts with the index Note: the EA principle (quantum-confined Stark effect) follows Miller et al. (1984); the ring modulator follows Xu et al. (2005). Note: curve shapes, shift sizes and ring size are schematic, not measured spectra. Note: dashed lines show the state with voltage applied.
Fig. 4 Conceptual diagram (vector drawing). The electro-absorption principle follows the paper by Miller et al. [Ref. 2]; that a ring modulator's resonant structure raises sensitivity to changes in refractive index follows the paper by Xu et al. [Ref. 7]. Curve shapes, shift widths and dimensions are schematic and do not show the spectrum of any particular device.

Electro-absorption (EA) — moving the absorption edge with a voltage

A semiconductor absorbs light at wavelengths shorter than a certain value and lets longer wavelengths through. Near that boundary (the absorption edge), applying an electric field to shift the edge lets you switch between passing and absorbing the light. In Physical Review Letters in 1984, Miller et al. showed in theory and experiment that applying an electric field perpendicular to the layers of a GaAs-AlGaAs quantum-well structure produces large shifts in optical absorption (the quantum-confined Stark effect)Sourced. Because the EA type does not rely on interference, it can be made short, and it lends itself to being combined on one chip with a laser made from the same compound semiconductor (commentary).

Microring (MR) — using resonance to amplify a small change

In Nature in 2005, Xu et al. reported a modulator using a silicon ring resonator. They wrote that the resonant structure "enhances the sensitivity of light to small changes in refractive index of the silicon", and demonstrated a modulator 12 µm in diameter, three orders of magnitude smaller than previous demonstrationsSourced. Because the wavelength of the resonance dip shifts with refractive index, the ring acts as a light-and-dark switch for light of a fixed wavelength (commentary). The catch is that the resonant wavelength also moves with temperature, a problem covered in our silicon photonics explainer.

TypeWhat it changesTypical sizeStrengthsWatch out for
Mach-ZehnderPhase (turned into light and dark by interference)Long (tends to be millimetre-scale)Phase control, QAMLength, drive voltage, nonlinear transfer characteristic
Electro-absorptionAbsorptionShortIntegration with the laserWavelength and temperature dependence of absorption
MicroringResonant wavelengthVery small (e.g. 12 µm diameter)Small size, low capacitance, wavelength multiplexingResonance shifts with temperature

"12 µm diameter" is Sourced from the paper by Xu et al. [Ref. 7]. The other cells are this article's framing based on physical principles (commentary), not a performance comparison of individual products.

6. A materials engineer's view (1): the refractive index barely moves

How much must the refractive index change to shift the phase by π in a Mach-Zehnder modulator? The phase shift is Δφ = 2π × Δn × L / λ (Δn: change in refractive index, L: arm length, λ: wavelength), so setting Δφ = π gives Δn = λ / (2L) (commentary).

Index change needed to shift the phase by π: Δn = λ/(2L) (our calculation) Calculated for λ = 1.31 µm. Bar length is proportional to Δn (1.31×10⁻³ = 440 px) L = 0.5 mm L = 1 mm L = 2 mm L = 5 mm 1.31 × 10⁻³ 6.55 × 10⁻⁴ 3.28 × 10⁻⁴ 1.31 × 10⁻⁴ An index change of 1/10,000 to 1/1,000, by voltage alone, tens of billions of times a second Note: our calculation from Δφ = 2πΔnL/λ, simplified to ignore optical confinement, push-pull drive and electrode efficiency. Note: this does not show the Δn achievable in any particular material or device.
Fig. 5 Drawing that includes our calculation (vector drawing). The phase formula is a standard optical relation; λ = 1.31 µm is an assumption chosen to match the nominal 1310 nm wavelength of the LPO specification [Ref. 3]. Each Δn was calculated by this article and is not a performance figure for any particular material or device.
Why this matters for materials engineers: modulator design is a game of making up for a small Δn with length or resonance

To shift the phase by π with a 1 mm arm takes a refractive index change of Δn = 1.31 µm / (2 × 1,000 µm) = about 6.6 × 10−4Our calculation. Raising and lowering the refractive index by a few parts in ten thousand, by voltage, tens of billions of times a second — that is the modulator's job.

When a material can deliver only a small Δn at a time, there are only two ways to make up the difference.

  • With length (Mach-Zehnder): longer arms lower the Δn required. But longer electrodes add electrical capacitance and loss and are harder to drive at high speed. Wang et al. say they solved this by simultaneously designing for "group-velocity matching" between the microwave and the lightSourced
  • With resonance (microring): the light circulates many times round the ring, so a small change per round trip accumulates. The circumference of Xu et al.'s 12 µm ring is π × 12 µm = about 38 µmOur calculation, a few tens of times shorter or more than a millimetre-scale MZ arm. The price is that it works only at one wavelength and is sensitive to temperature

Put another way, raise a material's electro-optic strength by an order of magnitude and the modulator can be an order of magnitude shorter. A shorter device has less electrode capacitance, and lower drive voltage and power. That is why new modulator materials such as thin-film lithium niobate and polymers continue to be researched (our commentary).

7. A materials engineer's view (2): the material decides how the effect works

Why this matters for materials engineers: "changing the refractive index" rests on entirely different physics in each case

The three material systems differ in the very mechanism by which they change refractive index or absorption, and that sets the character of each device (our commentary).

  • Lithium niobate (the Pockels effect): an electric field slightly displaces the ions in the crystal, and the refractive index changes in proportion to the field. The response tends to be free of absorption and linear. On the other hand, as Wang et al. write, "difficulties in microstructuring lithium niobate" long held back integrationSourced
  • Silicon (carrier-induced index change): crystalline silicon has no Pockels effect, so the refractive index is changed by using a voltage to move electrons and holes in and out. Carriers also absorb light, so changing the phase also changes the loss (a well-known general property of the material). Even so, as Liu et al. note, compatibility with CMOS processing and monolithic integration with electronics are advantagesSourced
  • Compound-semiconductor quantum wells (electro-absorption): well structures a few nanometres thick are built in by crystal growth to engineer where the absorption edge sits. Miller et al.'s work used GaAs-AlGaAs quantum wellsSourced. Precision in layer thickness and composition translates directly into the usable wavelength and the depth of extinction

Seen from a materials supplier's side, choosing a modulator means choosing which physical property can be built in, by which manufacturing process, to what precision. The Pockels effect is set by crystal symmetry, the carrier effect by the doping profile and the quantum-confinement effect by epitaxial layer thickness. Though all are "modulators", the materials technologies they call for — single-crystal growth, ion implantation, MOCVD — belong to entirely separate lineages (our commentary).

And as Section 3 showed, PAM4 and QAM demand modulator linearity. Between the Pockels effect, whose response is inherently linear, and the carrier effect, in which absorption and phase move together, producing the same four levels or sixteen points is not equally easy (our commentary).

8. What is still hard

(1) Towards 400 Gb/s per lane

While 200 Gb/s per lane (PAM4 at 106.25 GBaud) is still on its way through standardisationSourced, the signalling for the next generation has not been decided. OIF held a "448Gbps Signaling for AI Workshop" (April 2025), and an IEEE P802.3dj task force document lists the start of work on 400 Gb/s signalling (a Call for Interest, CFI) as a planned itemSourced. Whether the baud rate will double again or more bits will go on each symbol — the next generation's modulation format had not been settled as of this article's researchNot yet confirmed.

(2) No single modulator material has won

As this article has shown, LiNbO₃, silicon and compound semiconductors each work by different physics and each has different weaknesses. Wang et al.'s paper says that integrated modulators based on silicon, InP and polymers have, because of intrinsic material limitations, "not yet been able to meet ... simultaneously" the requirementsSourced. Which material will become the mainstream of the next generation cannot be judged within the scope of this articleNot yet confirmed.

(3) What this article does not cover

The electro-optic coefficients of each material, and the drive voltage, bandwidth and power of particular products, are not given, because the primary sources this article could open and check did not provide them in a comparable form. Soref and Bennett's 1987 paper (IEEE Journal of Quantum Electronics) is known as a foundational reference on electro-optic effects in silicon, but because this article could not check its text, no figures from it are quoted.

The article in summary
  • Modulation formats differ in how many bits each symbol carries. NRZ carries 1, PAM4 2 and DP-16QAM 8 (commentary)
  • PAM4 halves the speed but cuts the gap to 1/3 (about −9.5 dB)Our calculation
  • 400ZR is DP-16QAM at 59.84375 GBaud, for 478.75 Gb/s in totalSourced
  • The three main modulator types are Mach-Zehnder (phase and interference), electro-absorption (shifting the absorption edge) and microring (resonance) (commentary)
  • Shifting the phase by π with a 1 mm arm needs a refractive index change of about 6.6 × 10−4Our calculation
  • A thin-film LiNbO₃ modulator showed CMOS-compatible drive voltage, 210 Gb/s and less than 0.5 dB loss (Nature 2018)Sourced
  • Choosing a modulator material means choosing between separate lineages of materials technology: crystal symmetry, doping and epitaxial growth (our commentary)

9. Glossary

Modulation
Varying the intensity, phase, polarisation and so on of light in step with an electrical signal so that it carries information.
Baud rate (Bd)
The number of symbols sent per second. 1 GBd is a billion symbols per second.
NRZ
A format that sends one bit per symbol using the presence or absence of light (two levels).
PAM4
A format that sends two bits per symbol using four brightness levels.
QAM
A format that creates many points from combinations of phase and amplitude. 16QAM carries four bits per symbol.
DP (dual polarisation)
Putting separate signals on two orthogonal polarisations to double capacity.
Mach-Zehnder modulator
A modulator that splits light in two, introduces a phase difference and recombines the paths so that interference produces light and dark.
Electro-absorption modulator
A modulator that uses an electric field to shift a semiconductor's absorption edge, so that light is absorbed or passed.
Microring modulator
A modulator that switches light by using the refractive index to shift the resonant wavelength of a small ring resonator.
Pockels effect
An electro-optic effect in which the refractive index changes in proportion to the electric field. Found in LiNbO₃ and similar crystals.
Quantum-confined Stark effect
The shift of the absorption edge when an electric field is applied to a quantum well. The principle behind electro-absorption modulators.
Direct detection
Reception that measures only the brightness of the light. Used with NRZ and PAM4.

10. References (primary sources)

  1. Wang C. et al. (Nature, 2018) "Integrated lithium niobate electro-optic modulators operating at CMOS-compatible voltages", Nature 562, 101–104. DOI: 10.1038/s41586-018-0551-y — doi.org
  2. Miller D. A. B. et al. (Physical Review Letters, 1984) "Band-Edge Electroabsorption in Quantum Well Structures: The Quantum-Confined Stark Effect", Phys. Rev. Lett. 53, 2173–2176. DOI: 10.1103/PhysRevLett.53.2173 — doi.org
  3. LPO MSA "100G-DR-LPO Revision 1.0", March 2025 (PDF) — lpo-msa.org
  4. OSFP MSA "Specification for OSFP Octal Small Form Factor Pluggable Modules, Rev 5.22", 9 August 2025 (PDF) — osfpmsa.org
  5. OIF "Implementation Agreement 400ZR (OIF-400ZR-03.0)", 8 October 2024 (PDF) — oiforum.com
  6. OIF "Implementation Agreement for 800ZR Coherent Interfaces (OIF-800ZR-01.0)", 8 October 2024 (PDF) — oiforum.com
  7. Xu Q. et al. (Nature, 2005) "Micrometre-scale silicon electro-optic modulator", Nature 435, 325–327. DOI: 10.1038/nature03569 — doi.org
  8. Liu A. et al. (Nature, 2004) "A high-speed silicon optical modulator based on a metal–oxide–semiconductor capacitor", Nature 427, 615–618. DOI: 10.1038/nature02310 — doi.org
  9. IEEE P802.3dj Task Force "Timeline Consideration", 9 December 2025 (PDF) — ieee802.org
  10. OIF "OIF Publishes Implementation Agreement for 112 Gb/s Retimed Transmitter Linear Receiver (RTLR) ...", 18 November 2025 (the site's events listing includes "448Gbps Signaling for AI Workshop – April 15-16, 2025") — oiforum.com

11. Claim-to-source audit

Claim in the textBasisLabel
That conventional LiNbO₃ modulators are "bulky, expensive, limited in bandwidth and require high drive voltages". That integration of LiNbO₃ was hampered by "difficulties in microstructuring lithium niobate". That integrated modulators in silicon, InP and polymers have not yet met the requirements simultaneously. That CMOS-compatible drive voltage, up to 210 Gb/s and on-chip optical loss below 0.5 dB were demonstrated. That group-velocity matching between microwave and light was designed simultaneouslyWang et al., Nature 562, 101 (2018)Reference 1 https://doi.org/10.1038/s41586-018-0551-ySourced
That applying an electric field perpendicular to the layers of a GaAs-AlGaAs quantum-well structure was shown in theory and experiment to shift optical absorption substantially (the quantum-confined Stark effect)Miller et al., Phys. Rev. Lett. 53, 2173 (1984)Reference 2 https://doi.org/10.1103/PhysRevLett.53.2173Sourced
That 100G-DR-LPO uses 53.125 GBd PAM4 at a nominal 1310 nm, up to 500 m over SMF. The provision for host-side non-linear compensationLPO MSA 100G-DR-LPO specificationReference 3 https://www.lpo-msa.org/files/live/sites/lpomsa/files/specs/LPO_MSA_Specification_v1p0_final.pdfSourced
That 1.6TAUI-8 is eight lanes of 224G-PAM4 at 106.25 GBaud, for 1.6 Tb/s in totalOSFP MSA specification Rev 5.22Reference 4 https://osfpmsa.org/assets/pdf/OSFP_Module_Specification_Rev5_22.pdfSourced
That 400ZR is single-carrier coherent DP-16QAM at 59.843750000 GBaud per polarisation, 478.750 Gbps in total. Uses including amplified DWDM up to 120 kmOIF 400ZR implementation agreement Rev 3.0Reference 5 https://www.oiforum.com/wp-content/uploads/OIF-400ZR-03.0.pdfSourced
That 800ZR is DP-16QAM at a nominal 118.203350603 Gbaud, targeting single-span, amplified DWDM over 80 to 120 kmOIF 800ZR implementation agreement Rev 1.0Reference 6 https://www.oiforum.com/wp-content/uploads/OIF-800ZR-01.0.pdfSourced
That the ring resonator structure "enhances the sensitivity of light to small changes in refractive index of the silicon". That a modulator 12 µm in diameter, three orders of magnitude smaller than before, was demonstratedXu et al., Nature 435, 325 (2005)Reference 7 https://doi.org/10.1038/nature03569Sourced
That silicon waveguide modulators had previously reached only about 20 MHz. That a modulation bandwidth exceeding 1 GHz was shown with a MOS capacitor structure. Compatibility with CMOS processingLiu et al., Nature 427, 615 (2004)Reference 8 https://doi.org/10.1038/nature02310Sourced
That a P802.3dj task force document lists a CFI on 400 Gb/s signalling as a planned itemIEEE P802.3dj task force documentReference 9 https://www.ieee802.org/3/dj/public/25_1209/dambrosia_3dj_01b_251209.pdfSourced
That OIF held a "448Gbps Signaling for AI Workshop" (15–16 April 2025) (as listed in the site's events listing)OIF websiteReference 10 https://www.oiforum.com/oif-publishes-implementation-agreement-for-112-gb-s-retimed-transmitter-linear-receiver-rtlr-electrical-and-optical-interface-advancing-energy-efficiency/Sourced
20 log10(1/3) = −9.5 dB. 106.25 × 2 = 212.5 Gb/s (212.5 GBd for NRZ). log216 = 4, 8 bits with two polarisations, 59.84375 × 8 = 478.75 Gb/s, 118.203 × 8 = 945.6 Gb/s. Δn for each length from Δn = λ/(2L) (λ = 1.31 µm). π × 12 µm = 38 µmOur calculation. Assumes four ideal, evenly spaced levels and a simplification that ignores optical confinement and electrode efficiency. Bit rates include FEC and other overheadOur calculation
The modulation format of the next generation (around 400 Gb/s per lane). The mainstream modulator material of the next generationStill under study; an outlook not settled as of this article's researchNot yet confirmed
General explanations of modulation basics (intensity, phase, polarisation, baud rate, MZ interference, the EA absorption edge, resonance). The principle-based cells in the comparison table of the three types. The property that carriers in silicon absorb light. The framing of making up for a small Δn with length or resonance. The reading that a stronger material effect allows a shorter device. Framing the three material systems as separate lineages of materials technology (single-crystal growth, ion implantation, MOCVD). The reading on the difference in linearity between the Pockels and carrier effectsGeneral explanations of the physics, and this article's framing and commentary based on published literature. Not views expressed by any of the researchers or bodiesCommentary
Electro-optic coefficients of each material; the drive voltage, bandwidth and power of particular products. Figures from Soref and Bennett (1987)Not stated in this article, because the primary sources it could open and check did not provide them in a comparable form, or their text could not be checkedCommentary
That Figs. 1, 3 and 4 are explanatory drawings, that Figs. 2 and 5 are drawings that include our calculation, and that the hero image is AI-generatedA note by this articleCommentary

Last updated 25 September 2026. Sources are limited to primary material (peer-reviewed papers, OIF implementation agreements, the LPO MSA and OSFP MSA specifications, and public IEEE 802.3 documents). Because the article includes general explanations of the physics and materials-design interpretations, those are marked as Commentary and kept separate from sourced fact. The electro-optic coefficients of each material and the drive voltage, bandwidth and power of particular products are not given, because comparable primary sources could not be confirmed. The next generation's modulation format and modulator materials are treated as not yet settled as of this article's research. All figures are for explanation. Figs. 1, 3 and 4 are vector drawings, Figs. 2 and 5 are vector drawings that include our calculation, and the hero image is AI-generated; none of them shows the structure, spectrum or dimensions of a real device.

🌐 Japanese