How Microsoft Just Built the Most Stable Qubit Ever

One year ago, Microsoft announced the

world's first topological cubit, the

Myerana 1. Now they're back with the

Myerana 2, claiming they have extended

the cubit lifetime thousand times higher

than what we were able to achieve with

Myana one.

>> An improvement so huge that Microsoft

has just cut their timeline to a

scalable quantum computer in half,

meaning we may be seeing them roll off

the production line as early as 2029.

>> Shut up and take my money.

>> I was invited a few months ago to see

the facilities where these chips are

being developed. And look, in my

experience, this sort of progress in the

field is basically unheard of. The

nearest analog that we have, Schulov's

law, which is Moore's law, lesserknown

cousin, says that cubic coherence time

doubles about once every year. A 1,000

times gain represents 10 years of

progress in just 12 months. I wanted to

know what changed, so I called up

Microsoft to ask.

>> We've seen a,000x improvement in our

cubits in the last year based on an

underlying material change. This is a

big claim that potentially rewrites the

entire timeline to a useful quantum

computer. But to understand where it

came from, how it works, and whether it

holds up, we need to start with the easy

stuff. How to build a quantum computer.

>> That's the same computer astronauts use

to do their taxes. The start of all

quantum computing stories is that normal

computers run on bits of zero or one.

But quantum computers operate on cubits

which instead of being confined to just

zero or one can exist in a superp

position occupying both states

simultaneously and so allow calculations

that would otherwise take conventional

computers thousands of years to

complete. That property has kept

physicists in their own superp position

of barely contained excitement but also

perpetual dismay of just how hard it is

to maintain these cubits for long enough

to do something useful with them.

>> The problem is that these are actually

all kind of delicate that that these

these effects superposition interference

entanglement can all be disrupted in

effect by interactions or even

entanglement with the environment. Every

cubit platform has its own specific

vulnerability, but the common thread is

that quantum information is

extraordinarily sensitive to any

interaction with the environment.

Whether that's heat, vibration, or

radiation, any disturbance can cause it

to collapse in a process called

decoherence. That is why most quantum

computer chips that you see are held in

one of these, a dilution refrigerator.

And this one is the one that I visited

inside Microsoft's lab in Copenhagen. It

looks broadly like a steampunk

chandelier, but it is one of the most

carefully controlled environments ever

engineered by humans. It creates a

vacuum that removes around 99.9999%

of the air molecules that might

otherwise bump into and interrupt a

quantum computation. It also reaches

temperatures as low as 50 ml, which is

0.05

degrees above absolute zero. That's over

50 times colder than deep space and a

temperature at which atoms basically

stop vibrating. The shrouding around it

protects against electromagnetism.

Things like RF frequencies from your

phone antenna or any magnetic fields or

signals that may negatively influence

the system. All in an attempt to keep

this quantum state preserved. But even

with all of this, most quantum

approaches struggle to make their cubits

last for more than a few hundred

milliseconds. So the team at Microsoft

asked, is there just a better way? The

underlying kind of wager you know that

we took in pursuing this path was there

are certain phases of matter you know

that we've engineered like these

topological phases of matter or topo

conductors we call them uh that have

this underlying property that's you know

and I use this analogy a lot like this

is a merous band right and you know I

took a piece of two-sided tape and you

know twisted and glued it together and

um that you know if I stretch it I you

know I squeeze it if I were to sit on it

it's still going to have one twist and

you have to do something kind of extreme

to undo this twist which is have to you

know rip it open, untwist it and you

know glue it back and so that was the

you know the idea that we took is that

you could have states of matter that

automatically passively have this

feature.

>> That is the idea behind topological

quantum computing. You weave quantum

information into a global property of

the system rather than storing it

locally. But what on earth do any of

those words even mean? And how do you

actually build something like that?

Microsoft's cubits are built from

superconducting semiconductor nanowire.

An incredibly thin wire just 35

nanometers wide made from a

semiconducting base where electrons are

moved around using electric field gates

just like those in a computer chip. Then

on top of that wire they laid down an

extremely thin layer of superconducting

material roughly 30 atoms thick. Then

the whole thing is cooled down to just

50 ml. And at that point, the electrons

in the superconductor stop moving

independently and instead pair up into

Koopa pairs, which are able to move

through the material with zero

electrical resistance and are what give

superconductors their superconducting

properties. Because the two materials

are perfectly joined together, the

paired frictionless behavior of the

superconductor on top leaks across the

boundary into the semiconductor below, a

phenomenon called the proximity effect.

Then by applying a magnetic field in

parallel to the superconducting

semiconductor wire, the material can be

moved into a state where the quantum

mechanical properties of the system

depend on its overall structure rather

than any local detail. This is called a

topological phase. In this phase,

something unusual happens. Each

individual electron is essentially split

in half and each half exists at either

end of the wire. This phase is called a

myurana zero mode. And these aren't real

particles that you can actually put in a

jar and think about. They are quasi

particles, collective ripples in the

material that essentially behave like a

single individual particle. That is

strange enough in its own right, but it

hasn't answered the question of where

exactly is the cubit. By linking two of

these wires together into an H shape

called a tetron, the cubit is encoded

into whether the system overall holds an

even or odd number of electrons in

total, a property called par. Because

par is a property of the whole tetron,

not just any single location. The

information doesn't just live at either

end of the wire, but across all four

myano zero modes simultaneously. This is

the whole idea behind a topological

cubit. It extends quantum information

across an entire area rather than in a

localized point. And that matters

because ordinary cubits fail the instant

that they experience interaction with

their environment. Here there is no

single point that the quantum

information is contained. To read or

corrupt this cubit, the environment

would need to nudge both ends of these

wires at exactly the same time. And the

probability of that happening by

accident is pretty vanishingly small.

Meaning that topological cubits have a

potential to last thousands of times

longer than conventional cubits. This

whole idea that the best way to protect

a cubit is to spread it everywhere all

at once appears in papers dating back to

1997. Microsoft picked this concept up

in 2004 when mathematician Michael

Freriedman wrote a letter directly to

Bill Gates making the case that this may

be the best way to build a quantum

computer. Even from there though, it

still took 21 years and some of the best

minds in physics to produce the world's

first working topological cubit.

>> What Myron one showed is that we could

make topological cubits that we could

engineer the topological phase and we

could uh you know we had this like sort

of like H-shaped cubit the tetron cubit.

What we saw is with myron 1 was we'd

measure it and then we'd see that it

would kind of there'd be an error would

occur. It would kind of flip uh on a

time scale of 1 to 10 milliseconds.

You'd see a flip and that's an error.

That's bad, right? You don't want that

to happen.

>> Here, the team at Microsoft had done

incredibly well, but still not well

enough. With 1 to 10 milliseconds

between errors and each cubit operation

taking around one microcond, you're

looking at roughly 10,000 operations

before the cubit loses its state. Fault

tolerant algorithms typically need

millions to billions of operations per

logical cubit. 10,000 is orders of

magnitude too short. So whilst Microsoft

did show they were on to something

interesting, it just wasn't working well

enough yet. And that is what brings us

on to their latest announcement. Before

we get there, I want to talk about a

problem I'm having. As you might be able

to tell, I put a lot of thought into

working out exactly how I want to

explain an idea. And I do all of this by

having a lot of conversations. The

problem is once the words leave my mouth

oftent times they leave my brain too.

That is exactly what is being solved by

today's sponsor Plaude and their note

pin S which is actually something I use

to make content on this channel. It is a

wearable AI notetaker. You can clip it,

pin it or wear it on your wrist and I

have it in almost every single

brainstorming session I do whilst I try

and understand ideas before coming

decoherent myself. You can press record.

The conversation then happens and when

it ends the plaude app produces a

transcript, structured notes, a summary

and a to-do list automatically. It runs

for 20 hours on a single charge, which

is a fact that I have tested myself,

which is impressive and also worrying

about how many meetings that I do in a

single day. Something I find really

useful is this button that you can see

here. You can press it mid-con

conversation to flag a moment that you

feel is essential. You can also just ask

Plaude and it has a built-in AI

assistant to retrieve information. I

asked it to pull together every point

from a producer call the week before,

and it came back with the decisions and

enough context that I didn't need to

relisten to a single second. For

interviews, producer meetings, or

anything involving multiple people

talking over each other, I also use the

Plaude Note Pro. It snaps magnetically

to the back of your phone, picks up

audio across 16 ft with four

microphones, and separates overlapping

voices. It does the same workflow as

before, but now with perperson notes, a

structured summary and analysis of where

the conversation landed and what was

left unresolved. Both are built around

enterprise level security and privacy

standards. And it's worth saying here,

tell people that you hit record button

and check that it is legal in the

country that you live. If you'd like to

check out PLA, the link is down in the

description. They are currently running

the biggest sale of the year, so use the

code Ben 15 for the current promotional

pricing, and there is also a 30-day free

return if it's not for you. Thank you,

Plaude, for sponsoring the channel. Now,

back to the video.

On the 2nd of June this year, last week,

as I'm filming this, Microsoft announced

that they had increased the cubit

lifetime from 12 milliseconds to over 20

seconds, a 1,000 times improvement, with

some cubits even lasting for over a

minute. That improvement came from

solving a problem the topological

architecture can't solve on its own.

Because despite its protection strategy,

there are still two ways the myana cubit

can be destroyed. The first is something

called quasi particle poisoning. In a

superconductor, electrons bind into

Koopa pairs. But if something in the

environment, for example, a stray photon

carries enough energy to rip one of

those pairs apart, you get a free

electron wandering through your system.

If it either joins the wire or leaves

the wire, it can flip the parity of the

cubit from 0ero to one or vice versa.

The energy needed to break apart a cubit

is called the superconducting gap and it

is different for every material. Through

the proximity effect, the semiconductor

inherits a fraction of this gap from the

superconductor above. This inherited

energy barrier is called the topological

gap. It is always smaller than the

superconducting gap, but the two scale

together. So if you make the

superconducting gap bigger, the

topological gap grows with it. The

second failure route is myana

hybridization. The two myorana zero

modes at opposite ends of the wire

aren't perfectly isolated. There is a

tiny quantum mechanical coupling between

the two of them. And this small energy

splitting causes the cubit to wobble

slightly away from zero energy. That

wobble is its own error channel. Both

failure modes here trace back to the

same root cause, which means that the

fix in principle was straightforward.

Find a superconductor with a bigger gap.

The Myerana 1 used aluminium or

aluminum. the obvious first choice

because it's exceptionally well

understood material that is compatible

with semiconductor fabrication and has

predictable properties. But aluminium

superconducting gap is around 300

microeleron volts. And to put that in

perspective, a photon of infrared

radiation, the kind emitted by anything

warm, including the walls of the

refrigerator itself, carries enough

energy to exceed that. As a result,

Microsoft swapped out the aluminium

superconducting layer for one that was

made of lead, which has a

superconducting gap of around 1,300

microeleron volts. This means that

breaking up the Koopa pairs now takes

four times the amount of energy than in

the Myerana 1.

>> Really big improvements. And very

importantly, it's in a very specific

predictable way, right? Which is that we

said all along, oh, we're just going to

make this gap larger and our cubid's

going to get better and better. The

obvious question here is why didn't they

just use lead from the start? That is

because the superconductor doesn't just

sit on top of the semiconductor in any

loose sense. It has to be bound to it at

an almost impossibly precise level.

Every interface in that material stack

has to be essentially perfect with every

atom in the right place because at this

scale even a single misplaced atom could

be a cubit killer. During my visit to

Microsoft Quantum Lab, I got to see the

device where they actually grow these

crystals in one atomic layer at a time.

A machine so precise it can control the

position of every individual atom.

Getting it to work with lead in

particular, though, a material that is

notoriously contaminating and sticks to

basically every surface you don't want

it to, particularly in fabrication

equipment. And making it grow at low

temperature without disturbing the

semiconductor beneath or polluting the

machine for future runs of fabrication

took them years to figure out. But after

thousands of failed depositions and what

I imagine is an ungodly number of hours

spent cleaning contaminated chambers,

what they arrived at was this the Myana

2 chip that increased the topological

cubic lifetime of the Myerana 1 from 12

microsconds to now over 20 seconds.

>> So it's a really big difference and

actually it's very funny like if you

know sometimes I look at some of the

plots in our you know our recent paper

and uh they look kind of like the plots

from a year ago. The only difference is

the axis over here is minutes. You know,

>> to put that new number in context, for a

chip where each operation takes one

microscond, this means that the Myana 2

can perform 20 million operations before

any error has a realistic chance of

occurring. To quote Microsoft, the

probability of any unintended par flip

during a typical cubit operation becomes

effectively negligible. What this

actually means is that for the first

time, cubit lifetime isn't the problem

standing between us and a useful quantum

computer. made that gap larger and the

cube has got better and so by by the way

we remember how to do that. So in terms

of future advances we you know we know

the same levers apply now other

approaches I think don't have those

levers that we have and and that's why

they I don't think they can do these

things.

>> What remains is scaling getting four

cubits to behave this way is one thing.

Getting a thousand to behave this way

and preserving those properties as the

array gets larger and more complex is

the engineering challenge that is left

to conquer. The question is, how long

does that take?

>> It was uh do you have a computer nearby?

>> As part of their Myana 2 announcement,

Microsoft also announced that they now

believe that they can cut their

timelines to a useful quantum computer

in half, now targeting 2029. I asked how

confident Chaitton actually was on this

new revised 2029 schedule.

>> It's it's two things actually. We have

this path forward, which is we make the

topological gap better by making these

materials improvements and the cubis

would get better. That was great idea on

paper. We were very confident in it. But

now we've proved that out, right? We

have now we we've shown that actually we

can deliver on that promise. The second

thing that I think is very important is

in the last year or two there have been

a lot of algorithmic improvements in

terms of how many cubits and how long

it'll take you to solve some key

problems. Like take take short shores

algorithm for instance if you want to

break RSA 2048 a year and a half ago

people would have said well you probably

need around 10 million cubits and it

came down to more like a million and

then 800,000 then now it sounds like

it's maybe like 600,000. So so the

target is moving and it's moving closer

right so you've got the requirements

coming down this way and and the

hardware going this way and so we're

really asking like when are these things

going to intersect and now they're both

moving towards each other. So this

progress to a useful quantum computer

seems to have several levers that are

pulling it in the right direction. The

thousandfold improvement in cubit

lifetime in just one year as a direct

consequence of material change is a very

impressive physics result. I don't think

that it necessarily means that we have a

new moors law of cubits just yet. The

superconducting gap has physical limits

and there will be material swaps and

improvements that can be made in the

future I imagine. But eventually

Microsoft will just have to find other

levers to keep making progress. I do

also always want to mention here in

breakthroughs like this there will be

problems and this one is no different.

There has always been some level of

controversy around Microsoft's

topological cubit program. They redacted

an early paper for this technique after

push back from the physics community

that their measurements of the Myerana

zero mode weren't sufficiently

definitive. And there is definitely some

grumbling going on with this

announcement too that the evidence that

Microsoft has published isn't sufficient

to back up their claims. And right now

there is no demonstration of things like

superp position control, no gate

operations or any algorithm executions.

And it is reasonable to comment that

this longived state is not the same

thing necessarily as a perfectly

controllable cubit and we haven't been

shown it in action. The response to all

of this is that Microsoft has

unpublished data showing full cubit

control and algorithm runs on chip and

they're being released slowly and that

the runner 2 is a representative move to

being more transparent. Whether that is

enough to move the skeptics is a

separate question, but we're just going

to have to watch and see how this

conversation actually evolves. Whether

2029 or not is the correct exact year, I

will be honest. I'm probably not at the

point where I want to put any money on a

specific date, but what happened here is

a significant leap in the right

direction. What I will say is that

having spent real time getting my head

around the result and having had the

chance to sit down with Chaitton, the

physics and the engineering on display

here are extraordinary. The sheer

precision required at every level of

this system, from the atomic interfaces

I saw demonstrated in Copenhagen to the

topological protection strategy of this

cubit, is pushing the limit of human

innovation in an unbelievably

complicated field. One thing I didn't

have time to cover in this video is how

Microsoft is actually reading out

quantum information from the Myeron

estates without destroying them. Jaden

gave me a really great explanation of

this and it, as well as the ad-ree

version of this video, is available on

Patreon and in the YouTube members

section. This was a really cool

announcement. Some people might even

call it a quantum leap. Thank you so

much for watching. I will see you next

time. Goodbye. And for the first time

ever, our computer lab actually has a

computer in it.