TECHNOLOGY EXPLAINER
IOWN
— a network that never turns light back into electricity, and devices that carry light into the computer
IOWN (pronounced "eye-own") is NTT's vision for a communications and computing infrastructure built around light. The headline numbers are 100 times the power efficiency, 125 times the transmission capacity and 1/200 of the end-to-end latency, but these are targets set for 2030. How far have the All-Photonics Network (APN), launched in 2023, and PEC-2, a photonics-electronics convergence device that links boards with light, actually come? This article separates the language of the vision from the substance of the technology.
- What IOWN is (the short version)
- Three targets, and the roadmap
- APN: inside a network that never goes back to electricity
- Our calculation: latency inside the fiber cannot be reduced
- PEC devices: four stages by distance
- A materials engineer's view (1): optical parts that drop hermetic sealing and survive reflow
- PEC-2: linking boards with light
- PEC-3 and PEC-4: into the package
- A materials engineer's view (2): a substrate that removes heat set the speed of light modulation
- Open problems and unconfirmed points
- Glossary / References / Claim-to-source audit
Sourced = stated in published material from the NTT Group or a public body (link given)
Our calculation = a figure this article derived from assumptions it spells out
Not yet confirmed = vision targets, commercialization plans and other items with no confirmed track record yet
Structural readings and materials-design interpretations are marked separately as commentary.
IOWN's "100 times, 125 times and 1/200" are targets of the vision and are written separately from performance already achieved. Company figures are labelled as commercial service specifications, prototype values, demonstration measurements or targets, as the case may be.
1. What IOWN is (the short version)
IOWN (Innovative Optical and Wireless Network) is, in NTT's words, a vision for creating a richer society using cutting-edge optical technologySourced. NTT here is the Japanese telecoms group formerly named Nippon Telegraph and Telephone, which changed its trade name to NTT, Inc. in July 2025. IOWN is not the name of a specific product but an umbrella term for a plan to rebuild both networks and computers around light (our commentary).
- Three technologies: NTT names the technologies that make up IOWN as the All-Photonics Network (APN), Photonics-Electronics Convergence (PEC) and the AI Computing Platform (AICP)Sourced
- What it changes: in the network, communication stays as light from end to end, without ever being converted back to electricity (APN). In the computer, electrical wiring is replaced with optical waveguides to cut the power spent on interconnects (PEC)Sourced
- Where it stands: NTT says IOWN has already started commercial service, and that it aims for widespread adoption around 2030Sourced. The timing of widespread adoption is Not yet confirmed
The general case for co-packaged optics (CPO), which places optical engines in the switch package, OIF's implementation agreement for 3.2T modules and the standard for external laser sources were covered in our explainer on co-packaged optics. Silicon photonic components and coupling are covered in our silicon photonics explainer, multicore fiber in our optical fiber explainer, and coherent transmission and DSP in our explainer on coherent optical communications and DSP. This article narrows the focus to what is specific to NTT's IOWN: how APN works, the technology in each PEC generation, and the distinction between targets and results (our own division of topics).
2. Three targets, and the roadmap
Three numbers come up in every account of IOWN. For APN, NTT writes that it aims to achieve its target performance for 2030 (100 x power efficiency, 125 x capacity, 1/200 end-to-end latency)Sourced.
| Target | Stage at which NTT says it will be reached | Label in this article |
|---|---|---|
| End-to-end latency of 1/200 | NTT says that with IOWN1.0 (the launch of the APN service) it achieved the latency target of 1/200. But the NTT EAST announcement attaches a condition: latency for video traffic within the same prefecture that no longer needs compression | NTT's claim (with conditions)Sourced |
| 125 times the transmission capacity | Aimed for with IOWN3.0 | Not yet confirmed |
| 100 times the power efficiency | Aimed for with IOWN4.0 | Not yet confirmed |
Based on NTT's IOWN "Functions and Features" page [Ref. 1], NTT's APN page [Ref. 2] and an NTT EAST press release [Ref. 3]. The baseline for comparison is given as conventional electronics-based communication networks [Ref. 3].
The most important thing in this figure is that with each generation, the distance light covers gets shorter: starting with 1000 km telecom networks, then 10 m between boards, centimeters between packages and millimeters between dies. NTT explains that as transmission capacity (frequency) rises, electrical signalling comes to consume a great deal of energy even over short distances, so that it becomes worthwhile to replace electricity with light even inside the computerSourced.
3. APN: inside a network that never goes back to electricity
NTT describes the problem with today's networks as followsSourced. In packet communication (TCP/IP), data is split into packets with destinations attached, and every router in the network checks the destination each time. Even if optical fiber is used for transmission, the signal has to be switched back to electricity every time.
APN rethinks this, completing communication entirely in light from one end to the other, without ever converting back to electricity. By eliminating data splitting and electrical conversion, NTT says it delivers an ultra-low-latency networkSourced. Technically, it is a wavelength network that provides end-to-end optical wavelength paths from the terminal through the networkSourced.
On 16 March 2023, NTT EAST and NTT WEST, NTT's regional carriers for eastern and western Japan, began providing the first commercial service, "APN IOWN1.0"Sourced. The announcement said the following.
| Item | What NTT EAST announced |
|---|---|
| Circuit | A dedicated, point-to-point 100 Gbps circuit. An OTU4 interface was added to the existing 100 Gbps high-speed broadband access service |
| Features | Latency 1/200 of the conventional level (latency for video traffic within the same prefecture that no longer needs compression) and zero jitter. Because the optical wavelength is dedicated, other users' traffic has no effect |
| Terminal equipment | "OTN Anywhere": receives 10GbE and 100GbE optical signals and converts them to OTU4. It measures and visualizes latency between sites and can adjust it |
| Distance unit for pricing | Where the distance between the serving buildings exceeds 35 km, charges apply per 35 km |
All Sourced (NTT EAST press release, 2 March 2023 [Ref. 3]).
On the capacity side, too, NTT has been building up experiments. It reports transmission at 1 Tbit/s per wavelength with 35 wavelengths multiplexed in the laboratory in 2019, and a field demonstration in 2023 of 336 km transmission at 1.2 Tbps per wavelength over installed commercial fiber, and says it is aiming for transmission of around 1 Pbit/s per fiber using multicore fiber and other meansSourced. The 1 Pbit/s class is a Not yet confirmed target. Splicing technology for multicore fiber was covered in our optical fiber explainer.
4. Our calculation: latency inside the fiber cannot be reduced
- Assumptions: a group index of 1.47 for silica fiber (the same assumption as in our optical fiber explainer) and a speed of light of about 300,000 km/s
- Propagation delay in fiber is about 4.9 µs per km Our calculation
- About 0.17 ms for the 35 km pricing unit and about 0.49 ms for 100 km (one way)
This propagation delay is a physical floor set by how fast light travels through glass, and no change of network architecture shrinks it (our commentary). Given that NTT EAST attaches video traffic that no longer needs compression as a condition for 1/200, what 1/200 mainly reduced can be read as the time spent on electrical processing along the way (compression, packet processing, queuing and so on) (our commentary).
Assumptions and limits: the group index is an assumption made in this article. The breakdown and absolute value of latency in the conventional service could not be confirmed from primary sources within the scope of this article, so no specific values for the numerator and denominator of 1/200 are given.
5. PEC devices: four stages by distance
NTT calls the technology that integrates electronic and optical circuits to build compact, highly efficient optical transceiver functions photonics-electronics convergence, and refers to the resulting family of devices as PEC (Photonics-Electronics Convergence) devicesSourced.
| Generation | What it links, and distance | Commercialization timing (as stated) | Concrete form (as stated) |
|---|---|---|---|
| PEC-1 | Between data centers (1000 km to 100 km, telecom) | CY2021, CY2023 | COSA, CoPKG. Introduced in the digital coherent optical transceivers for APN IOWN1.0 |
| PEC-2 | Between boards (about 10 m, computing) | CY2025 to CY2026 | Photonics-electronics convergence switch (CPO). 3.2T, then 6.4T |
| PEC-3 | Between packages (cm scale) | CY2028 | Optical chiplet. Commercial samples planned for the fourth quarter of 2028 |
| PEC-4 | Between dies (mm scale) | CY2032 | — |
Based on NTT Innovative Devices' briefing material (6 October 2025) [Ref. 4] and the February 2026 NTT Technical Journal [Ref. 5]. Existing PEC-1 and PEC-2 products and prototypes are Sourced; the commercialization timing of PEC-2 and all of PEC-3 and PEC-4 are Not yet confirmed. Note that the November 2023 NTT Technical Journal [Ref. 6] calls COSA the "second generation" and the optical engine for CPO the "third generation": the generation count differs between documents. This article follows the PEC-1 to PEC-4 naming of [Refs. 4 and 5].
6. A materials engineer's view (1): optical parts that drop hermetic sealing and survive reflow
The November 2023 NTT Technical Journal gives an explanation, interesting from the materials side, of the COSA (coherent optical subassembly) developed for long-haul digital coherent optical transceiversSourced.
- All optical circuits other than the light source are integrated on a single silicon photonics chip (about 4 mm × 6 mm), in a package of 13.5 mm × 10.5 mm × 2.2 mm
- What made the dramatic miniaturization compared with earlier discrete optical devices possible was being able to omit the temperature-control section and the package not needing to be hermetic
- That was the result of exploiting the material stability of silicon while achieving temperature-independent characteristics and moisture resistance through NTT's own optical circuit design
- Because the package is non-hermetic, the fiber end face could be connected directly to the silicon photonics chip
- With the fiber bonded directly in place, NTT achieved a structure and materials that withstand solder-reflow mounting temperatures (about 250 °C), allowing the part to be mounted automatically by BGA together with other electronic components. COSA has been commercial since 2020
For a long time, optical components were sealed in hermetic metal or ceramic packages, with a Peltier element holding the temperature constant (our commentary). COSA did away with both. What it demands instead is optical circuits whose characteristics do not change when directly exposed to moisture, and a fiber joint that stays in position and stays transparent through a 250 °C reflow.
The optical path coupling adhesives rated for 260 °C reflow and the partitioning of adhesive functions described in our explainer on optical connectors and fiber attach are exactly the materials answers to this requirement. From protecting parts with their environment, through hermetic sealing, to protecting them through the moisture and heat resistance of the materials themselves: IOWN's devices stand on that shift (our commentary).
7. PEC-2: linking boards with light
PEC-2 is a device that converts between electricity and light right next to the switch chip inside a data center switch. The February 2026 NTT Technical Journal describes an OIF-compliant 3.2 Tbit/s PEC-2 and a prototype "photonics-electronics convergence switch" that carries it, as followsSourced.
| Item | As stated in the NTT Technical Journal (3.2T version, prototype) |
|---|---|
| Configuration | Modulators and receivers densely integrated in silicon photonics, with a DSP converting between electricity and light. 32 channels, 3.2 Tbit/s in total, supporting 100 Gbit/s PAM4 signals |
| Compared with pluggables | Equivalent to eight 400G pluggables. Module size reduced by 87%. Power consumption 48 W, 50% less than eight 12 W pluggables |
| Prototype switch | PEC-2 mounted next to a 51.2 Tbit/s switch chip. Front panel of 64 16-fiber MPO connectors, 2RU, water cooling in addition to air cooling |
| Field operation | At Expo 2025 Osaka, Kansai, from 13 April to 13 October 2025, it completed six months of stable operation as part of a real-time AI video analysis system |
| Issue | Because PEC-2 is soldered next to the switch chip, the equipment has to be stopped and taken apart if it fails |
All Sourced (NTT Technical Journal, February 2026 [Ref. 5]). These are prototype values, not the specification of a volume-production product.
For the commercial version, NTT Innovative Devices set out the following plan at its October 2025 briefingNot yet confirmed.
- A 6.4T optical engine and a 102.4 Tbps CPO switch module, with about 50% power reduction for the switch alone
- A socketed optical engine, so that it can be removed and replaced even though it is CPO (a fundamental cut in the cost of repairs after failures)
- Partners are Broadcom (the 102.4 Tbps switch LSI) and Accton (the switch chassis). Optical engine samples in Q2 2026 and commercial CPO switch samples in Q4
- Performance targets of 3.9 pJ/bit and 0.4 Tbps/mm (bandwidth density per unit edge length). Manufacturing capacity is being expanded to 5,000 units per month per line
NTT Innovative Devices' briefing material [Ref. 4]. All are plans or targets; at the time of writing (September 2026), this article had not confirmed from primary sources that commercial samples have shipped.
- Eight pluggables: 12 W × 8 = 96 W ÷ 3.2 Tbit/s = about 30 pJ/bit Our calculation
- PEC-2 (3.2T prototype): 48 W ÷ 3.2 Tbit/s = about 15 pJ/bit
- The 3.9 pJ/bit target for the 6.4T optical engine: 6.4 Tbit/s × 3.9 pJ/bit ≈ about 25 W (per engine)
- The 0.26 pJ/bit PEC-3 device demonstration: 64 Gbit/s × 0.26 pJ/bit ≈ about 17 mW (per lane)
Assumptions and limits: each measures a different scope. The 96 W and 48 W cover the whole module, including maximum values under the OIF specification; 3.9 pJ/bit is a target for the optical engine; 0.26 pJ/bit is a demonstration value for the devices alone, such as the laser, modulator and photodetector (Section 8). They are not compared on the same basis.
8. PEC-3 and PEC-4: into the package
PEC-3 is an "optical chiplet" that links packages such as GPUs to each other with light. NTT Innovative Devices' material states the followingSourced.
- Commercial samples planned for the fourth quarter of 2028; prototypes are under evaluationNot yet confirmed
- The target for drastically lower power consumption is about 0.26 pJ/bit
- In demonstrations with a configuration including a DFB laser, an electro-absorption modulator (EAM) and a photodetector: 0.26 pJ/bit at 64 Gbit/s NRZ (laser current 10 mA) and 0.14 pJ/bit at 50 Gbit/s (5 mA)
- At its core is the NTT laboratories' own "membrane device": a structure with a very thin active layer in which light travels inside the III-V semiconductor. The diagram shows a combination of an InP taper and a SiOx waveguide
Under NEDO's "Research and Development Project of the Enhanced Infrastructures for Post-5G Information and Communication Systems / Development of Advanced Semiconductor Manufacturing Technologies" (NEDO is Japan's New Energy and Industrial Technology Development Organization, a government R&D funding agency), the organizations for three themes under "packaging technology for photonics-electronics convergence" were selected in January 2024Sourced.
| R&D theme | Type | Lead proposer | Co-proposers |
|---|---|---|---|
| Optical chiplet packaging technology | Commissioned | NTT | NTT Innovative Devices, Furukawa Electric, NTT Device Cross Technologies, Shinko Electric Industries |
| Photonics-electronics convergence interface memory module technology | Commissioned | Kioxia | NTT |
| Deterministic-latency computing infrastructure technology | Subsidized | NTT | NEC, Fujitsu |
All Sourced (NTT news release, 30 January 2024 [Ref. 7]; NEDO's decision on the implementation structure [Ref. 8]). According to NEDO, the project period is in principle up to five years from the start of R&D for the two commissioned themes and up to three years for the subsidized theme.
9. A materials engineer's view (2): a substrate that removes heat set the speed of light modulation
The membrane technology said to be at the heart of PEC-3 includes an example of a material choice raising the performance ceiling. In October 2020, NTT and Tokyo Institute of Technology developed an InP-based membrane laser on a SiC substrate and reported a 3 dB bandwidth of more than 100 GHz as a directly modulated laser (published in Nature Photonics)Sourced.
A membrane laser is a thin-film laser about 300 nm thick (roughly a tenth of a conventional one) built on a low-index material. NTT explains that because a low-index material lies directly beneath it, light and carriers are strongly confined in the active layer, which promises low power consumption and high speedSourced.
Until then, however, that low-index material had been thermal oxide (SiO2). NTT says that because SiO2 conducts heat poorly, the active-layer temperature rose sharply when current flowed, the gain saturated and the relaxation oscillation frequency levelled off at around 20 GHzSourced. In other words, the material used to confine the light was blocking the exit for heat (our commentary).
The solution was a SiC substrate. NTT says SiC has about 500 times the thermal conductivity of SiO2 and a lower refractive index than InP, so optical confinement stays almost the same as on SiO2Sourced. The release's glossary gives SiC a thermal conductivity of 490 W/(m·K) and a refractive index of 2.58Sourced. The breakthrough was meeting, in one material, a combination that is normally hard to get together: low refractive index and high thermal conductivity (our commentary).
The description of the process will also be familiar to bonding engineersSourced.
- SiC and InP were directly bonded through 40 nm of SiO2 using oxygen plasma. The SiO2 combines bond strength with no damage to the active layer, but since it conducts heat poorly it is made as thin as possible
- To suppress film delamination caused by gas released when the temperature is raised after bonding, grooves to let the gas escape were formed in the substrate beforehand
- By calculation, going from 2 µm to 40 nm of SiO2 lowers the temperature rise of the active layer from 130.9 °C to 16.8 °C (assuming a 50 µm active layer and 100 mW of heat)
Bond-line thickness, outgassing and thermal conduction: the leading-edge performance of photonics-electronics convergence turns out to be decided by the materials and process fundamentals of wafer bonding and thermal design (our commentary). How this relates to the "SiOx waveguide" and "InP taper" in the PEC-3 material could not be confirmed in detail from primary sources within the scope of this article.
10. Open problems and unconfirmed points
(1) Most of the targets are still ahead
125 times the capacity (IOWN3.0) and 100 times the power efficiency (IOWN4.0) are targets that NTT itself says it is "aiming for"Not yet confirmed. NTT says the 1/200 latency has been achieved, but with the condition that it applies to video traffic within the same prefecture that no longer needs compressionSourced. It does not mean latency falls to 1/200 for every kind of communication (our commentary).
(2) PEC-2 commercialization is a 2026 plan
For the 6.4T optical engine and the 102.4 Tbps CPO switch, the plan is optical engine samples in Q2 2026 and commercial samples in Q4Not yet confirmed. At the time of writing (September 2026), this article had not confirmed from primary sources that commercial samples have shipped.
(3) PEC-3 and PEC-4 are at the research stage
PEC-3's 0.26 pJ/bit is a device-level demonstration value, not a value for an optical chiplet built into a GPU packageSourced. Commercial samples are planned for the fourth quarter of 2028, and PEC-4 for 2032Not yet confirmed.
(4) What this article does not cover
Specifications from the IOWN Global Forum (the international promotion body), details of the AI Computing Platform (AICP), and the latest pricing and service areas for APN were left outside the scope of this article.
- IOWN is a vision. 100 times the power efficiency, 125 times the capacity and 1/200 of the latency are targets for 2030Not yet confirmed
- APN IOWN1.0 launched commercially in March 2023. It is a dedicated 100 Gbps circuit, and the 1/200 comes with the condition of video traffic within the same prefecture that no longer needs compressionSourced
- Propagation delay in fiber (about 4.9 µs/km) does not shrink with the architecture. What 1/200 reduced can be read as electrical processing along the wayOur calculation
- The PEC-2 3.2T prototype uses 48 W (50% less than eight pluggables). Commercial samples of the 6.4T version are planned for 2026Sourced
- Towards optical parts that drop hermetic sealing and survive reflow. COSA omitted temperature control and hermetic sealing, and achieved a structure and materials that withstand 250 °C reflowSourced
- Moving to a SiC substrate raised the bandwidth ceiling of the membrane laserSourced
11. Glossary
- IOWN
- Innovative Optical and Wireless Network. NTT's vision for a communications and computing infrastructure centered on light.
- APN
- All-Photonics Network. A network that connects end to end with optical wavelength paths and does not convert back to electricity along the way.
- Optical wavelength path
- An optical route in which one wavelength is dedicated from sender to receiver.
- OTU4
- An interface of about 100 Gbps class in the optical transport network (OTN) standardized by ITU-T.
- Photonics-electronics convergence (PEC)
- Technology that integrates electronic and optical circuits to build compact, highly efficient optical transceiver functions.
- COSA
- Coherent optical subassembly. A component combining the optical circuits with the drivers and TIAs in one package.
- CPO
- Co-packaged optics. Mounting optical engines right next to the switch chip.
- Optical engine
- The non-pluggable optical transceiver section used in CPO and similar designs.
- pJ/bit
- The energy used to send one bit. The smaller, the lower the power.
- Optical chiplet
- A small photonics-electronics convergence chip for linking packages with light. NTT's PEC-3.
- Membrane laser
- A thin-film laser about 300 nm thick built on a low-index material.
- Relaxation oscillation frequency
- The frequency that sets the upper limit on speed when a laser is modulated directly by its current.
- Direct bonding
- Joining surface-treated materials to each other without adhesive.
- Non-hermetic package
- A package whose interior is not completely sealed. The materials themselves must resist moisture.
- Solder reflow
- The heating step that solders all the components on a board at once. About 250 to 260 °C.
12. References (primary sources)
- NTT "IOWN Functions and Features" (explanation of APN, PEC, AICP and the roadmap; Japanese page) https://group.ntt/jp/group/iown/function/
- NTT "APN | Functions and Features | IOWN" (Japanese page) https://group.ntt/jp/group/iown/function/apn.html
- NTT EAST "Launch of APN IOWN1.0", press release, 2 March 2023 (in Japanese) https://www.ntt-east.co.jp/release/detail/20230302_01.html
- NTT Innovative Devices Masahito Tomizawa, "Development status and outlook of the photonics-electronics convergence switch toward IOWN 2.0", 6 October 2025 (NTT IR briefing material, PDF, in Japanese) https://group.ntt/jp/ir/library/presentation/2025/pdf/251006_2.pdf
- NTT Technical Journal Yusuke Muranaka et al., "A high-capacity, low-power photonics-electronics convergence switch equipped with PEC-2", February 2026 (in Japanese) https://www.rd.ntt/research/JN202602_38182.html
- NTT Technical Journal Shin Kamei and Yuzo Ishii, "Photonics-electronics convergence devices supporting IOWN (development of second- and third-generation devices)", November 2023 (in Japanese) https://journal.ntt.co.jp/article/23720
- NTT "NTT selected as an implementing company for NEDO's Post-5G Information and Communication Systems Infrastructure Enhancement R&D Project", 30 January 2024 (in Japanese) https://group.ntt/jp/newsrelease/2024/01/30/240130a.html
- NEDO "Decision on the implementation structure for the Post-5G Information and Communication Systems Infrastructure Enhancement R&D Project / Development of Advanced Semiconductor Manufacturing Technologies", 30 January 2024 (in Japanese) https://www.nedo.go.jp/koubo/IT3_100302.html
- NTT "World's fastest directly modulated laser with a bandwidth exceeding 100 GHz developed: achieved with low power consumption by a membrane laser on a SiC substrate", 20 October 2020 (in Japanese) https://group.ntt/jp/newsrelease/2020/10/20/201020a.html
- NTT "What is IOWN?" (Japanese page) https://group.ntt/jp/group/iown/whats.html
13. Claim-to-source audit
| Claim in the text | Basis | Label |
|---|---|---|
| The three technologies making up IOWN (APN, PEC, AICP). That electrical signalling consumes a great deal of energy even over short distances as capacity rises. That each router in packet communication converts back to electricity, and that APN is completed in light end to end without converting back. The two approaches of AICP and that PEC replaces electrical wiring with optical waveguides. NTT's statement that IOWN1.0 launched the APN service and achieved the 1/200 latency target, that IOWN2.0 is a board-connected photonics-electronics convergence device, and that IOWN3.0 aims for 125 times capacity and IOWN4.0 for 100 times power efficiency | Reference 1 https://group.ntt/jp/group/iown/function/ | Sourced |
| That APN is a wavelength network providing end-to-end optical wavelength paths. The target performance for 2030 (100 x power efficiency, 125 x capacity, 1/200 latency). The 2019 experiment of 1 Tbit/s × 35 wavelengths, the 2023 field demonstration of 1.2 Tbps over 336 km, and the aim of about 1 Pbit/s per fiber. The launch of APN IOWN1.0 on 16 March 2023 | Reference 2 https://group.ntt/jp/group/iown/function/apn.html | Sourced |
| The launch date of APN IOWN1.0, the dedicated point-to-point 100 Gbps circuit, OTU4, latency 1/200 of the conventional level (for video traffic within the same prefecture that no longer needs compression), zero jitter, dedicated wavelength, the functions of OTN Anywhere, and the 35 km pricing unit. That the baseline is conventional electronics-based communication networks | Reference 3 https://www.ntt-east.co.jp/release/detail/20230302_01.html | Sourced |
| The connection targets, distances and Commercialization years for PEC-1 to PEC-4 (CY2021 and 2023, CY2025–2026, CY2028, CY2032). The 6.4T optical engine and 102.4T CPO switch module, 50% power reduction for the switch alone, the socketed optical engine, Broadcom and Accton, the Q2 to Q4 2026 plan, 5,000 units per line per month, and targets of 3.9 pJ/b and 0.4 Tbps/mm. PEC-3 commercial samples planned for Q4 2028, the target of about 0.26 pJ/bit, demonstrations of 0.26 pJ/bit at 64 Gbit/s (10 mA) and 0.14 pJ/bit at 50 Gbit/s (5 mA), and the membrane device | Reference 4 https://group.ntt/jp/ir/library/presentation/2025/pdf/251006_2.pdf | Sourced |
| The definition of photonics-electronics convergence technology and PEC devices. The introduction of PEC-1 in APN digital coherent optical transceivers with IOWN1.0. The configuration of the PEC-2 3.2T prototype (32 channels, 100G PAM4, DSP, silicon photonics), 87% size reduction, 48 W (50% less than eight 12 W units), the 51.2 Tbit/s switch, 64 16-fiber MPOs, 2RU, air plus water cooling, six months of operation at the Expo, the difficulty of replacement because of soldered mounting, and consideration of providing commercial-version samples in 2026 | Reference 5 https://www.rd.ntt/research/JN202602_38182.html | Sourced |
| COSA's silicon photonics chip of about 4 mm × 6 mm, the 13.5 × 10.5 × 2.2 mm package, omission of the temperature-control section and non-hermeticity, temperature independence and moisture resistance, direct connection of the fiber end face, a structure and materials that withstand reflow at about 250 °C, BGA, and commercialization since 2020. That COSA is called the "second generation" and the optical engine for CPO the "third generation" | Reference 6 https://journal.ntt.co.jp/article/23720 | Sourced |
| The three themes of the NEDO project (optical chiplet packaging technology, photonics-electronics convergence interface memory module technology, deterministic-latency computing infrastructure technology), their lead and co-proposers, and whether commissioned or subsidized | Reference 7 https://group.ntt/jp/newsrelease/2024/01/30/240130a.html | Sourced |
| That NEDO decided the implementation structure on 30 January 2024, the theme categories, and the project periods (in principle up to five years and up to three years) | Reference 8 https://www.nedo.go.jp/koubo/IT3_100302.html | Sourced |
| The InP-based membrane laser on SiC (with Tokyo Institute of Technology, in Nature Photonics). 3 dB bandwidth above 100 GHz (108 GHz); relaxation oscillation frequency 42 GHz and 3 dB bandwidth 60 GHz. Saturation at around 20 GHz on SiO2. SiC's thermal conductivity about 500 times that of SiO2, almost equal optical confinement, SiC at 490 W/(m·K) and index 2.58. Film thickness about 300 nm. Direct bonding through 40 nm SiO2 with oxygen plasma, outgassing grooves, and the calculated temperature rise of 130.9 °C to 16.8 °C | Reference 9 https://group.ntt/jp/newsrelease/2020/10/20/201020a.html | Sourced |
| The definition of IOWN (a vision for creating a richer society using cutting-edge optical technology), that commercial service has already started, and the aim of widespread adoption around 2030 | Reference 10 https://group.ntt/jp/group/iown/whats.html | Sourced |
| Achieving 125 times capacity and 100 times power efficiency; widespread adoption around 2030; PEC-2 commercial samples (2026); PEC-3 (2028) and PEC-4 (2032); about 1 Pbit/s per fiber | NTT Group targets and plans, not results. Reference 4 https://group.ntt/jp/ir/library/presentation/2025/pdf/251006_2.pdf | Not yet confirmed |
| Propagation delay (about 4.9 µs/km, about 0.17 ms for 35 km, about 0.49 ms for 100 km). About 30 and about 15 pJ/bit; 6.4T × 3.9 pJ/bit ≈ about 25 W; 64 Gbit/s × 0.26 pJ/bit ≈ about 17 mW | Our calculation. The group index of 1.47 and a speed of light of about 300,000 km/s are this article's assumptions. The pJ/bit values each measure a different scope and are not compared under the same conditions | Our calculation |
| The reading that 1/200 mainly reduced electrical processing along the way. The framing of a shift from hermetic sealing to protection by the materials themselves. The framing that SiO2 was blocking the exit for heat. The mapping between IOWN1.0 to 4.0 and the PEC generations. The framing that IOWN is an umbrella term for a plan rather than a product name. The general explanation that conventional optical components used hermetic packages and temperature control | This article's commentary based on published content. Not views expressed by NTT | Commentary |
| The absolute value and breakdown of latency in the conventional service; a record of commercial PEC-2 sample shipments; details of the relationship between PEC-3's SiOx waveguide and the membrane | Not stated, because they could not be confirmed from primary sources within the scope of this article | Commentary |
| That Figs. 1, 2 and 5 are explanatory drawings and Fig. 3 a drawing that includes our calculation, and that the hero image and Fig. 4 are AI-generated images | Our note | Commentary |
Last updated 26 September 2026. Sources are limited to primary material (the NTT Group's official websites, press releases, IR briefing material and technical journal, and NEDO's award announcement); market estimates from research firms are not used. IOWN's "100 times the power efficiency, 125 times the capacity and 1/200 of the latency" are targets of the vision, and the 1/200 latency is something NTT says it has achieved under stated conditions. Everything from PEC-2 commercialization onwards is a plan. The absolute latency of the conventional service and a record of commercial PEC-2 sample shipments are not stated because they could not be confirmed in published primary sources. All figures are for explanation. Figs. 1, 2 and 5 are vector drawings, Fig. 3 is a vector drawing that includes our calculation, and the hero image and Fig. 4 are AI-generated images; none of them shows real equipment, a real device or a real cross-section.