The Fiber Problem: Why Distance Breaks Quantum Networks

Every quantum network built so far runs into the same wall: how do you move a quantum state across distance without destroying the property that makes it quantum in the first place. That's the fiber problem. 

Diagram shows the photon's path: native wavelength out, up-converted into the fiber band, lossy fiber crossing, down-converted back, with the no-cloning constraint called out separately underneath since it's the reason none of that loss can be patched mid-flight.

Classical data degrades over distance too, but repeaters fix it. A signal gets weak, a repeater reads it, copies it, and sends a fresh copy further down the line. Quantum states can't do that. The no-cloning theorem, a hard result in quantum mechanics, says you cannot create an identical copy of an unknown quantum state. There's no way to read a photon's quantum state, copy it, and pass along a clean version. Doing so destroys the very state you were trying to preserve. Whatever leaves the source has to survive, unassisted, all the way to the destination.

Fiber makes survival harder the farther a photon travels. Standard telecom fiber loses roughly 0.2 decibels of signal per kilometer at the wavelength commercial networks use, a small number that compounds fast over distance. A classical signal can be boosted past that loss. A single photon carrying a quantum state can't be boosted, only lost. Every kilometer of fiber is another chance for that photon to get absorbed or scattered before it arrives.

There's a second problem. Commercial fiber carries light efficiently only in narrow bands, mainly around 1550 nanometers, because that's where glass loses the least light over distance. The atomic systems used to store and process quantum information, trapped ions, neutral atoms, certain solid-state qubits, usually don't emit or absorb light at 1550 nanometers. A photon leaving an atomic memory typically needs a wavelength conversion step just to enter the fiber network, and another one on the far end to be absorbed by the receiving memory. Each conversion is one more place to lose or degrade the state.

You can't clone a quantum state to fix loss along the way, and fiber adds distance-dependent loss plus a wavelength mismatch on top of photons that are already fragile.

In my next post I'll describe how Brookhaven National Labs and Stony Brook University have worked around the fiber "problem."

This is the problem Chapter 1 of Quantum from the Ground Up opens on. You can read the current edition of the book here.

IBM Links Two Quantum Cryostats Toward Fault Tolerance

Chapter 3 of Quantum from the Ground Up makes one argument: the physics works, and the engineering catches up slowly, one specialized system at a time. The chapter leans on the dilution refrigerator as its example, a device that cools a quantum chip down to a few thousandths of a degree above absolute zero. A separate post here, The Chandelier, walked through why that cooling matters. Below a certain temperature, the metal on a quantum chip becomes a superconductor, meaning electricity flows through it with no resistance. That property is what lets a qubit hold information as a quantum state instead of losing it to heat and vibration almost instantly. IBM's announcement on August 19 is the next chapter in that same argument. Instead of one isolated refrigerator, two of them, joined.

What IBM Actually Did

On August 19, IBM announced it had physically connected two cryogenic modules—each its own standalone refrigerator—into a single integrated cold environment. Crucially, this initial milestone served as a structural, thermal, vacuum, and EMI-shielding validation of the empty joined cells, rather than a live multi-QPU execution. IBM plans to install its next-generation Nighthawk quantum processors into these coupled modules later in 2026 to execute live inter-module quantum gate operations.

A cryogenic module in this context is a sealed, vacuum-insulated box that removes heat in stages as you go deeper inside it, ending at a chamber cold enough to keep a chip in its quantum state. IBM reached a base temperature more than 180 times colder than deep space, which works out to below 15 millikelvin, or 15 thousandths of a degree above absolute zero. Getting there took five days at 4 Kelvin, the temperature of liquid helium, before a final drop to that base temperature shortly after. Standalone, each module stands about 8 feet tall and 8 feet wide, closer in size to an industrial appliance or plant assembly than a benchtop instrument.

The two modules IBM cooled down were empty, no chips installed. IBM plans to install Nighthawk quantum processors into the modules later this year, then test whether chips connected across the module boundary can run operations reliably enough to be useful. IBM itself has said that reliability across the L-coupler connection is still active engineering work, not a finished result. This week's announcement is the refrigerator working, not chips talking to each other across it.

Each module is a self contained casing that steps a chip down through three stages: room temperature electronics at the top, a 4 kelvin stage in the middle, then the chip itself below 15 millikelvin at the bottom. The line between the two chip stages is the L-coupler, the connection that lets Module A and Module B behave as one cooled environment instead of two separate refrigerators.

Why Wiring Is the Bottleneck

A quantum chip does not run itself. Every qubit on it needs a wire carrying a control signal in and a readout signal out, and those wires have to pass through every cooling stage without carrying stray heat down with them. That is the actual scaling problem, not the chip. IBM's new module design gives each vacuum enclosure up to 12 times more wiring space than today's most common IBM systems. More wiring space is what will let more chips be wired up and connected, both inside one module and across two joined modules, once IBM installs chips and runs them through the connector it calls the L-coupler. IBM describes the goal as processors reliably working on the same problem together, not just sitting side by side, which is a different requirement than just packing more qubits onto one chip.

Physical Qubits Versus Logical Qubits

This is also a good place to separate two terms I’ve used a little loosely. A physical qubit is one actual quantum circuit on a chip. A logical qubit is a group of many physical qubits, wired together and error corrected as a unit, that behaves like one reliable qubit for the purpose of running a calculation. Fault tolerant means a system can keep correcting its own errors fast enough to finish a long calculation before the errors pile up and ruin the answer.

IBM's roadmap calls for at least 1,000 programmable qubits by 2027 using L-couplers, feeding into IBM Quantum Starling, the system the company expects to deliver in 2029 as the first fault-tolerant quantum computer, running 100 million quantum gates across 200 logical qubits. Two hundred logical qubits sounds small next to 1,000 physical qubits, and that gap is the whole point. Error correction is expensive. However, while standard 2D surface codes typically require an overhead of 1,000+ physical qubits per logical qubit, IBM relies on quantum Low-Density Parity Check (qLDPC) codes. This reduces the overhead to roughly 50 physical qubits per logical qubit (~10,000 physical qubits for 200 logical qubits), making fault tolerance achievable on a substantially smaller hardware scale.

None of that works if the chips cannot be wired together and kept at the same ultra low temperature at the same time. The cryogenic housing announced this week is the part of the plan nobody puts on the cover of a press release, even though Live Science frames it as one of the field's biggest infrastructure bottlenecks. Without a way to link cryostats, a 200 logical qubit machine stays a slide in a roadmap deck.

What This Changes in the Book

Chapter 5 covers IBM's Condor chip: 1,121 physical qubits inside one standard cylindrical cryostat. That chapter treats the cryostat as fixed, one chip, one refrigerator. This announcement breaks that assumption. The box shaped modular design, not the cylinder, is now IBM's stated path past a single chip's qubit ceiling. Chapter 3's engineering argument gets a dated, measured example: five days to 4 Kelvin, then to base temperature, in a system built to add modules rather than grow one tank.

This post rolls into the next edition of the book, due September 1. The current edition is available at gordostuff.com/p/quantum-from-ground-up-hardware.html

A Gas Swap Fixes a Manufacturing Problem in Superconducting Qubits

Chapter 5 of Quantum From The Ground Up covers superconducting qubits: Josephson junctions, the Dolan bridge fabrication technique, and IBM's 1,121-qubit Condor chip running at 99.0 to 99.5 percent two-qubit fidelity. That chapter never asks how the metal underneath those junctions gets onto the chip in the first place. A new paper out of Cornell fixes a problem in that step, and it matters more than another fidelity number would.

A superconducting qubit is a tiny circuit built from metal that, once cooled near absolute zero, behaves like a single quantum object instead of an ordinary wire. The part that makes it a qubit rather than just a very cold wire is the Josephson junction: two superconducting metal layers separated by an insulating gap so thin that electrons tunnel straight through it. Tantalum has become a favorite metal for those layers because tantalum-based qubits hold their quantum state longer than most alternatives. But tantalum only works if its atoms land on the chip in one specific crystal arrangement, called the alpha phase. Getting that arrangement has required heating the substrate past 400°C during deposition. Most semiconductor factories run their fabrication lines with a hard ceiling near that same 400°C, so there was almost no margin between what tantalum needed and what a foundry could tolerate, as Cornell's team describes the manufacturing squeeze.

Cornell's group, led by Assistant Professor Valla Fatemi, builds these tantalum layers by sputtering: a process gas is ionized and fired at a block of tantalum, knocking atoms loose so they land on a silicon wafer and build up a thin film. The standard process gas is argon. Fatemi's team swapped in krypton, a heavier noble gas, and found that the heavier atoms push the tantalum into the alpha phase at temperatures as low as 200°C, half of what argon requires, according to the published results in Nature Materials on August 18, 2026. The resulting films also carried noticeably higher electronic conductivity than tantalum deposited the old way.

Krypton ions hit the tantalum target at the top, knock atoms loose, and those atoms travel down to build the alpha-tantalum film on the heated silicon substrate. Note 200°C, half the temperature argon sputtering needs.

The work builds on an earlier study from Fatemi's lab that sputtered niobium films with argon and mapped out how surface chemistry during deposition shapes final qubit performance. Swapping the metal to tantalum and the gas to krypton let the team apply what that niobium work taught them about controlling film quality. Qubits built from the new tantalum films performed, in Fatemi's own assessment, at the leading edge for the field.

None of this sets a new fidelity record. It removes a bottleneck that sits underneath every fidelity number in the chapter: a qubit design cannot scale to production if the fabrication step it needs falls outside what an ordinary chip factory can actually run.

This result covers one sample film deposited at one lab. Scaling it to production is a separate engineering problem, not a physics problem, and it is the harder one. A foundry running krypton sputtering needs the crystal phase, conductivity, and thickness to stay uniform across every chip on a wafer and across every wafer in a batch, not just in the sample that made it into the paper. It needs the process to hold up next to every other step already running on that line, since real chips stack tantalum with silicon oxide, aluminum wiring, and the Josephson junction itself, each with its own temperature limits and contamination risks. And it needs the equipment: krypton sputtering targets, gas handling, and chamber tuning are not yet standard equipment at most semiconductor fabs the way argon sputtering is. Lowering the temperature ceiling was the physics half of the problem. Building a repeatable, monitored, high-yield process around that lower ceiling, at the volume a real qubit chip production line runs, is the engineering half, yet to be figured out.

What This Changes in the Book

Chapter 5 lists IBM's Condor at 1,121 qubits and 99.0 to 99.5 percent two-qubit fidelity, built with Dolan-bridge Josephson junctions. This post adds a manufacturing footnote to the process. A lower-temperature tantalum deposition route now exists that fits inside standard semiconductor foundry limits. The path to building chips at that quality, at scale, gets wider.

This post will fold into the next edition of Quantum from the Ground Up, due September 1. The current edition is available on the book page.

What Mrs. Anderson’s High School Chemistry Class Taught Me About Units

I found a bunch of posts I started writing years ago but never finished. Here’s one of them.... just finished.

I remember Mrs. Anderson at the blackboard in my high school chemistry class, writing out a conversion problem - something like converting a volume in liters to milliliters to moles. She worked it by stacking fractions, one after another, each one arranged so a unit in the numerator of one fraction matched the unit in the denominator of the next. Then she crossed them out in pairs until only the answer's units remained. "Learn a few equations," she said, "and you can solve just about anything." She was not talking about memorizing formulas. She was talking about what she called factor labeling.

Factor labeling - today more commonly called a more fancy dimensional analysis - treats units as algebraic objects. You multiply and divide them the same way you multiply and divide numbers. A conversion factor like 1,000 milliliters per liter is really just the number one, with some units, so multiplying by it changes the label without changing the quantity. Chain enough of these factors together and the units in between cancel, leaving you with exactly the unit you wanted.

The value of the method is not speed. It is error detection. In high school it helped in chemistry (not my favorite subject). In college physics (loved it) it became something I depended on, once problems started combining velocity, acceleration, force, and energy in the same calculation and a single wrong exponent could hide inside an otherwise reasonable looking number..

Here’s a simple example. In circuit analysis, say a resistor carries 25 milliamps (I) at 12 kilohms ® and you need to figure the power (P) dissipated in watts. Run the raw numbers straight through P = I²R without tracking units and you get 7,500 - off by a factor of a thousand from the real answer. Run it through factor labeling instead: convert 25 milliamps to 0.025 amps and 12 kilohms to 12,000 ohms before multiplying, and the units confirm the answer lands in watts: 7.5 watts.

I have used this same check in electrical engineering courses, in circuit analysis, and in grading student calculations for capstone projects. A wrong answer with clean unit cancellation is rare. A wrong setup almost always leaves a stray unit sitting where it should not be.

Fifty plus years after high school, I always check my units before I trust my numbers. Thanks, Mrs. Anderson!

What Nobody Tells You About AI and Your Water Supply

Last week during a workshop on AI tools in the classroom at Pace University, a faculty member asked about AI and water use. I expected the question. The workshop was about classroom applications, not infrastructure, so I kept the answer short and moved on. Here’s a longer answer.

Data centers consume water directly, through cooling towers that evaporate it to remove heat. US data centers directly consume an estimated 17 to 19 billion gallons a year, a figure projected to climb toward 60 to 110 billion gallons by 2030. The bigger draw is indirect. Generating the electricity these facilities use consumed roughly 211 billion gallons of water in 2023, mostly through cooling at power plants. A data center's water footprint depends as much on how its power is generated as on how its servers are cooled.

Location compounds the problem. Two thirds of new data center construction since 2022 has landed in areas already under water stress. Municipal systems supply nearly all of that water, and the infrastructure upgrades to meet rising demand often fall on residential ratepayers rather than the companies building the facilities. Transparency has not kept pace either. Most operators disclose only partial or aggregate figures, so the public rarely sees facility level numbers.

The fixes already exist - sort of. Closed loop and liquid cooling systems cut water draw at the facility to near zero. Siting near hydro, nuclear, or wind power removes most of the indirect draw tied to electricity generation. The Massachusetts Green High Performance Computing Center in Holyoke, built in 2012 on hydro power from the Connecticut River, has been proving this model for over a decade. But.... scale matters here. MGHPCC runs on a 15 to 19 megawatt connection with about 10 megawatts dedicated to computing. A single hyperscale AI campus now under construction can draw a gigawatt or more, one hundred times that load. Holyoke proves the cooling and power sourcing approach works. It does not prove that approach scales directly to a facility built for AI though.

Regulation is starting to catch up. State legislatures introduced more than 200 data center bills in 2025 and enacted over 40 of them across 21 states. Most of what passed addressed ratepayer protection and siting incentives rather than water directly. Roughly 30 of the proposed bills targeted water consumption specifically, and only a handful of those were enacted. South Carolina and Kansas are considering mandates requiring closed loop cooling. California, Iowa, and Michigan are pushing water use disclosure requirements. Virginia is taking a different route, tying grant funding to the use of reclaimed wastewater instead of potable water in cooling systems. None of this is settled yet. Water legislation is still catching up to where energy legislation was two years ago, and disclosure requirements are starting to follow the path carbon reporting took a decade earlier.

I told that faculty member the honest version this week: the technology to fix this exists, and it works today in Holyoke. Scaling it up is an engineering problem, and engineers solve those. What is missing is the will to require it before the next facility breaks ground.

How Claude and Gemini Watermark Text, and Who Can Read It Back

You've probably heard Anthropic started watermarking Claude's text output on August 2, 2026, weaving an invisible mark into generated text and attaching signed provenance metadata to supported files. The move follows Anthropic signing the EU AI Act's Article 50(2) Code of Practice on Transparency of AI-Generated Content, which pushes providers toward making AI output machine detectable. Google got there first. Its SynthID system has been marking Imagen, Veo, Gemini, and audio output since 2023, and Google reported more than 10 billion pieces of content watermarked by May 2026. Anthropic's move gives a useful reason to look at how Google's SynthID method works and compare it to what Anthropic is doing.

Older systems tagged AI content with a label in the file's metadata. That never held up. Anyone could strip a label with a screenshot or by copying text into a new document. SynthID takes a different approach. It builds the mark into the content itself while the AI creates it.

For images, SynthID (used by Google Gemini) makes tiny changes to colors across the whole picture, changes too small for a person to notice. Those changes are spread out rather than placed in one corner, so cropping or compressing the image does not erase them.

For text, SynthID works differently, because there are no pixels to adjust. Every time the AI chooses the next word in a sentence, it usually has several reasonable options. SynthID tips that choice slightly, using a token sampling method that favors certain words over others in a pattern only its own detector can recognize. Read the sentence and it looks completely normal. The pattern lives in the pattern of word choices, not in any single word.

SynthID embeds the mark during generation. Detection returns a confidence score, not a yes or no.

Anthropic's Claude approach runs on a separate track entirely. When a supported Claude model writes a response, it weaves its own imperceptible watermark directly into the text as it generates. Anthropic says the mark does not change the meaning, quality, or readability of the response, and because it lives in the text itself rather than in a metadata tag, it travels along when the text is copied and pasted and may survive some editing. For files such as SVG, PNG, and JPG, Claude attaches signed C2PA provenance metadata instead, a different mechanism built for a different job. C2PA metadata is cryptographically signed and publicly verifiable, so anyone with the right tool can check it. A text watermark is statistical rather than signed, so checking it requires Anthropic's own detector, which the company has said it will publish along with technical documentation. Neither mark proves who wrote the underlying ideas. A watermark can appear on text that a person wrote and then ran through Claude for editing or translation, and its absence does not prove a human wrote something either.

SynthID and Claude's watermark work on the same principle. Neither system reads the other's mark.

A detector, trained alongside the system that creates the marks, scans content and reports how confident it is that SynthID is present, rather than a flat yes or no. That detector only works on content made by a model that has SynthID built in. It cannot identify AI content from a system that never used it, and heavy editing or rewriting can weaken the mark enough to make it hard to detect.

Access to that detector differs by content type and by company. Google open sourced the SynthID text detector through Hugging Face and GitHub, so any developer can run it against their own text. Image and video detection stays inside Google's own products, through tools like the Gemini app and Search. Anthropic has not published a Claude detector yet. The company has said it plans to release one along with technical documentation, but until that happens, checking a piece of text for Claude's watermark means trusting Anthropic's own tools rather than an independent one. SynthID's open text detector and Claude's still unpublished one sit at opposite ends of that spectrum, even though both watermarks work on the same basic principle.

My Stake In This - I've written here for almost 22 years and published five books along the way. Watermarking protects what a model outputs. It does nothing for what went into training it. I am confident that a good amount of my own writing sits somewhere in the data these companies trained on, unlabeled and uncredited, collected long before anyone built a watermark to mark the difference between a machine's words and mine.

Ten billion watermarked items sounds like a lot. It still only covers one company's models. Anthropic uses its own watermark for Claude rather than SynthID, and Apple's Image Playground has its own approach too. The EU AI Act is pushing every major provider toward some version of this, but each one is building a separate system with a separate detector. 

Post-Quantum Crypto Just Got a Chip to Run On

Chapter 14 of Quantum from the Ground Up covers the hardware side of post-quantum cryptography, meaning new encryption methods designed to survive an attack from a future quantum computer. This week gave that a concrete data point. BTQ Technologies and Taiwan's Industrial Technology Research Institute validated the first phase of a chip architecture built specifically to run the new encryption standards in hardware instead of software.

Why Hardware Matters Here

Encryption, whether old or new, is math. A processor runs that math the same way it runs any other program: by fetching instructions and data from memory, doing the calculation, then writing the result back. Post-quantum encryption methods use larger keys and more complex math than the encryption in use today, so they demand more of that fetch-and-calculate cycle. Running them purely in software, on a general-purpose processor, is slower and draws more power. That is a real problem for a car's onboard computer, a factory sensor, or a battery-powered IoT device that cannot spare the extra milliseconds or milliwatts.

The fix is to build a dedicated piece of hardware that runs only the encryption math, wired directly for that job. That is what a chip architecture like QCIM is.

What QCIM Actually Does

QCIM stands for Quantum Compute-in-Memory. Despite the name, it does not involve quantum computing itself. It refers to where the calculation happens on the chip. In a standard chip layout, memory and the processor are separate blocks connected by a data bus, and every calculation means shuttling data back and forth across that bus. Compute-in-memory design instead performs the calculation inside or immediately next to the memory itself, cutting out most of that back-and-forth. Less movement means lower power draw and faster results, which is exactly what power-constrained devices need if they are going to run demanding post-quantum math.

BTQ built this particular version around three specific encryption standards published by NIST: FIPS 203, 204, and 205. These are the official post-quantum algorithms the US government has approved for general use, covering both encrypting data and verifying digital signatures. Any hardware built to accelerate post-quantum cryptography needs to run these three algorithms specifically, since they are what systems will actually be required to support going forward.

 

What Was Tested, and What It Showed

The test ran inside a TSMC 28-nanometer design environment. TSMC is the world's largest chip manufacturer, and 28-nanometer refers to the size of the transistors in the manufacturing process being simulated, an older and well-proven node rather than the cutting edge, which keeps early testing cheaper and more predictable. Inside that environment, researchers checked two things. First, whether the QCIM core could genuinely speed up FIPS 203, 204, and 205 operations. Second, whether it produced correct results while doing so, since a faster chip that gets the math wrong is worthless for security. Both held up under the demanding conditions BTQ and ITRI put it through. ITRI's Dr. Chih-Cheng Lu called it meaningful progress toward module-level integration, the next phase of the program, where the core gets built into a larger working system rather than tested on its own.

BTQ is not new to this work. The collaboration with ITRI traces back to 2022, and a companion program with South Korea's ICTK is aimed at a fully integrated, commercially deployable chipset built around the same core. BTQ has said it expects to ship QCIM test chips to customers and partners by the end of the year. The target applications read like a list of things you do not want to re-secure one device at a time later: military systems, industrial equipment, automotive platforms, IoT devices, and connected infrastructure generally, all of which stay in service for years and are hard to patch remotely.

What This Changes in the Book

Chapter 14 already covers hardware paths to PQC, citing the SEALSQ QS7001. QCIM adds a second, independently developed example aimed at the same FIPS standards, this time using compute-in-memory design. The next edition adds it alongside QS7001 and tracks the module-level integration phase BTQ and ITRI are moving into next.

QCIM is the third hardware entry in a pattern this blog has been tracking since December: SEALSQ's QS7001, covered in The Quantum Security Race: Software vs. Hardware, and STMicroelectronics' ST54M, covered in Government Sets New Deadline for Quantum-Safe Encryption. All three chips target the same FIPS 203/204/205 standards introduced in Quantum Computers Just Got Much Closer to Breaking Your Passwords, and all three exist because the federal migration deadlines covered in that June post do not leave software-only implementations enough runway.

This post will fold into the next edition of Quantum from the Ground Up, due September 1.