0:00

to MIT Technology Review

0:02

Narrated. My name is Matt

0:04

Honan. I'm our editor -in -chief. Every

0:06

week, we'll bring you a fascinating, new,

0:09

in -depth story for the leading

0:11

edge of science and technology,

0:13

covering topics like AI, biotech,

0:15

climate, energy, robotics, and more.

0:18

Here's this week's story. I hope you enjoy

0:20

it. Narrated by

0:22

NOAA. Listen to more of

0:25

the best articles from the

0:27

world's biggest publishers on the

0:29

NOAA app. or at newsoveraudio.com.

0:33

Michael Brooks writes, quantum computing

0:35

is taking on its biggest

0:37

challenge. Noise. In

0:41

the past 20 years, hundreds

0:43

of companies, including giants like Google,

0:45

Microsoft and IBM, have

0:48

staked a claim in the rush to

0:50

establish quantum computing. Investors

0:52

have put in well over five billion

0:54

dollars so far. All this

0:56

effort has just one purpose, creating

0:58

the world's next big thing.

1:01

Quantum computers use the counterintuitive

1:04

rules that govern matter at

1:06

the atomic and subatomic level

1:08

to process information in ways

1:10

that are impossible with conventional

1:12

or classical computers. Experts

1:15

suspect that this technology will be able

1:17

to make an impact in fields

1:19

as disparate as drug discovery, cryptography,

1:22

finance, and supply chain

1:24

logistics. The

1:26

promise is certainly there. But so

1:28

is the hype. In

1:30

2022, for instance, Him

1:32

Israel, managing director of research at

1:34

Bank of America, declared that

1:36

quantum computing will be bigger than

1:38

fire and bigger than all the

1:41

revolutions that humanity has seen. Even

1:43

among scientists, a slew of claims

1:45

and vicious counterclaims have made

1:47

it a hard field to assess.

1:50

Ultimately, though, assessing our

1:53

progress in building useful quantum

1:55

computers comes down to one

1:57

central factor, whether we can

1:59

handle the noise. The

2:01

delicate nature of quantum systems

2:03

makes them extremely vulnerable to

2:05

the slightest disturbance, whether

2:07

that's a stray photon created by

2:09

heat, a random signal

2:11

from the surrounding electronics, or

2:13

a physical vibration. This

2:16

noise wreaks havoc, generating errors

2:18

or even stopping a quantum

2:20

computation in its tracks. It

2:23

doesn't matter how big your processor is,

2:25

or what the killer applications might turn

2:27

out to be, unless noise

2:29

can be tamed, a quantum

2:31

computer will never surpass what a

2:33

classical computer can do. For

2:36

many years, researchers thought they might

2:38

just have to make do with noisy

2:40

circuitry, at least in the near

2:42

term, and many hunted for

2:44

applications that might do something useful

2:46

with that limited capacity. The

2:49

hunt hasn't gone particularly well, but

2:51

that may not matter now. In

2:54

the last couple of years, theoretical

2:56

and experimental breakthroughs have enabled researchers

2:58

to declare that the problem of

3:00

noise might finally be on the

3:02

ropes. A combination

3:04

of hardware and software strategies

3:06

is showing promise for suppressing,

3:08

mitigating, and cleaning up quantum

3:10

errors. It's not an

3:12

especially elegant approach, but it does

3:14

look as if it could work.

3:17

and sooner than anyone expected. I'm

3:20

seeing much more evidence being presented

3:22

in defense of optimism," says Earl

3:24

Campbell, vice president of quantum

3:26

science at River Lane, a quantum

3:28

computing company based in Cambridge, UK. Even

3:31

the hard -line skeptics are being

3:33

won over. University of

3:35

Helsinki professor Sabrina Maniskalko,

3:37

for example, researches the

3:39

impact of noise on computations. A

3:42

decade ago, she says she was

3:44

writing quantum computing off. I

3:46

thought there were really fundamental issues. I

3:49

had no certainty that there would be a

3:51

way out," she says. Now,

3:53

though, she is working on using

3:55

quantum systems to design improved versions

3:57

of light -activated cancer drugs that

4:00

are effective at lower concentrations and

4:02

can be activated by a less

4:04

harmful form of light. She

4:06

thinks the project is just two and a

4:08

half years from success. From

4:10

Manus Galco, the era of

4:12

quantum utility, the point at

4:14

which, for certain tasks, it makes sense

4:17

to use a quantum rather than a

4:19

classical processor, is almost upon us. I'm

4:21

actually quite confident about the fact that

4:23

we will be entering the quantum utility

4:26

era very soon, she says. This

4:29

breakthrough moment comes after more

4:31

than a decade of creeping

4:33

disappointment. Throughout the

4:35

late 2000s and the early

4:37

2010s, Researchers building and running

4:39

real -world quantum computers found them

4:41

to be far more problematic

4:43

than the theorists had hoped.

4:46

To some people these problems

4:48

seemed insurmountable, but others, like

4:50

J. Gambetta, were unfazed. A

4:53

quiet -spoken Australian, Gambetta

4:55

has a PhD in physics

4:57

from Griffith University, on Australia's

4:59

Gold Coast. He chose to

5:01

go there in part because it allowed him

5:04

to feed his surfing addiction. But

5:06

in July 2004, He

5:08

wrenched himself away and skipped off to

5:10

the Northern Hemisphere to do research

5:12

at Yale University on the quantum properties

5:14

of light. Three years

5:16

later, by which time he was an

5:18

ex -surfer thanks to the chilly waters

5:20

around New Haven, Gambetta

5:22

moved even further north to the

5:24

University of Waterloo in Ontario, Canada. Then

5:27

he learned that IBM wanted to

5:29

get a little more hands -on with

5:31

quantum computing. In 2011,

5:34

Gambetta became one of the company's new

5:36

hires. IBM's quantum

5:38

engineers had been busy building

5:40

quantum versions of the classical

5:42

computer's binary digit, or bit. In

5:45

classical computers, the bit is

5:48

an electronic switch, with two states

5:50

to represent zero and one. In

5:53

quantum computers, things are less

5:55

black and white. If

5:57

isolated from noise, a quantum

5:59

bit, or qubit, can exist

6:01

in a probabilistic combination of

6:03

those two possible states. A

6:06

bit like a coin in mid -toss.

6:09

This property of qubits, along

6:11

with their potential to be

6:13

entangled with other qubits, is

6:15

the key to the revolutionary

6:17

possibilities of quantum computing. A

6:19

year after joining the company,

6:22

Gambetta spotted a problem with IBM's

6:24

qubits. Everyone could see that

6:26

they were getting pretty good. Whenever

6:28

he met up with his fellow physicists

6:30

at conferences, they would ask him

6:32

to test out their latest ideas on

6:34

IBM's qubits. Within a couple of

6:36

years, Gambetta had begun to balk at

6:38

the volume of requests. I

6:41

started thinking that this was insane. Why

6:43

should we just run experiments for

6:45

physicists, he recalls. It

6:47

occurred to him that his life might

6:49

be easier if he could find a

6:51

way for physicists to operate IBM's qubits

6:53

for themselves, maybe via cloud

6:56

computing. He mentioned it

6:58

to his boss. and then he found

7:00

himself with five minutes to pitch the

7:02

idea to IBM's executives at a gathering

7:04

in late 2014. The

7:06

only question they asked was whether Gambetta

7:08

was sure he could pull it off.

7:11

I said yes, he says. I

7:13

thought, how hard can it be? Very

7:16

hard, it turned out, because

7:18

IBM's executives told Gambetta he had

7:20

to get it done quickly. I

7:23

wanted to spend two years doing it, he

7:25

says. They gave him a year. It

7:28

was a daunting challenge. He

7:30

barely knew what the cloud was back

7:32

then. Fortunately, some of his

7:34

colleagues did, and they were

7:36

able to upgrade the team's remote access

7:38

protocols, useful for tweaking the machine

7:40

in the evening or on the weekend,

7:42

to create a suite of interfaces that could

7:45

be accessed from anywhere in the world. The

7:48

world's first cloud access

7:50

quantum computer, built using five

7:52

qubits, went live at

7:54

midnight on May 4, 2016.

7:57

The date, Star Wars Day,

7:59

was chosen by nerds

8:01

for nerds. I don't think

8:03

anyone in upper management was aware of

8:05

that, Gambetta says, laughing. Not

8:07

that upper management's reaction to the

8:09

launch date was uppermost in his mind.

8:12

A far more concern, he says, was

8:14

whether a system reflecting years of

8:17

behind -the -scenes development work would survive

8:19

being hooked up to the real

8:21

world. We watched the

8:23

first jobs come in. We could see

8:25

them pinging on the quantum computer,

8:27

he says. When it didn't break,

8:29

we started to relax. Cloud

8:32

-based quantum computing was an

8:34

instant hit. 7 ,000 people

8:36

signed up in the first week, and

8:38

there were 22 ,000 registered users by

8:40

the end of the month. Their

8:43

ventures made it clear, however, that

8:45

quantum computing had a big problem.

8:47

The field's eventual aim is to have

8:50

hundreds of thousands, if not millions,

8:52

of qubits working together. But

8:54

when it became possible for researchers

8:56

to test out quantum computers with just

8:58

a few qubits working together, many

9:01

theory -based assumptions about how much

9:03

noise they would generate turned out

9:05

to be seriously off. Some

9:07

noise was always in the cards. Because

9:10

they operate at temperatures above

9:12

absolute zero, where thermal radiation

9:14

is always present, everyone expected

9:16

some random knocks to the

9:18

qubits. But there were

9:20

non -random knocks, too. Changing temperatures

9:22

in the control electronics created

9:25

noise. Applying pulses of

9:27

energy to put the qubits in the

9:29

right states created noise. And

9:31

worst of all, it turned

9:33

out that sending a control signal

9:35

to one qubit created noise in

9:37

other nearby qubits. You're

9:39

manipulating a qubit and another one

9:41

over there feels it, says Michael Beersik,

9:44

director of the Quantum Control Laboratory

9:46

at the University of Sydney in Australia.

9:49

By the time quantum algorithms were running

9:51

on a dozen or so qubits, the

9:53

performance was consistently shocking. In

9:56

a 2022 assessment, Beersick

9:58

and others calculated the probability

10:00

that an algorithm would run successfully

10:02

before noise destroyed the information

10:05

held in the qubits and forced

10:07

the computation off track. If

10:09

an algorithm with a known correct

10:11

answer was run 30 ,000 times, say, the

10:14

correct answer might be returned

10:16

only three times. Though

10:18

disappointing, it was

10:20

also educational. People learned

10:22

a lot about these machines by actually

10:25

using them, Beersick says. We

10:27

found a lot of stuff that more or

10:29

less nobody knew about, or they knew

10:31

and had no idea what to do about it. Once

10:35

they had recovered from this

10:37

noisy slap, researchers began to

10:39

rally, and they have now

10:41

come up with a set of solutions that

10:43

can work together to bring the noise under control.

10:46

Broadly speaking, solutions can be

10:48

classed into three categories. The

10:51

base layer is error suppression. This

10:53

works through classical software and

10:55

machine learning algorithms, which continually

10:57

analyze the behavior of the

10:59

circuits and the qubits, and

11:02

then reconfigure the circuit design and

11:04

the way instructions are given so that

11:06

the information held in the qubits

11:08

is better protected. This is

11:10

one of the things that Beersick's

11:12

company, Q -Control, works on. Suppression,

11:14

the company says, can make

11:16

quantum algorithms 1 ,000 times more

11:18

likely to produce a correct answer.

11:22

The next layer, error

11:24

mitigation, uses the fact that

11:26

not all errors cause a computation

11:28

to fail. Many of them will

11:30

just steer the computation off track. By

11:33

looking at the errors that noise

11:35

creates in a particular system running

11:37

a particular algorithm, Researchers can apply

11:39

a kind of anti -noise to

11:41

the quantum circuit to reduce the

11:43

chances of errors during the computation

11:45

and in the output. This

11:47

technique, something akin to the

11:49

operation of noise -canceling headphones, is

11:52

not a perfect fix. It relies,

11:54

for instance, on running the algorithm

11:56

multiple times, which increases the cost

11:58

of operation, and the algorithm

12:00

only estimates the noise. Nonetheless,

12:03

it does a decent job of

12:05

reducing errors in the final output, Gambetta

12:07

says. Helsinki -based Algorithmic,

12:09

where Maniscalco is

12:11

CEO, has its own way

12:13

of cleaning up noise after the computation is

12:15

done. It basically eliminates

12:17

the noise in post -processing, like

12:19

cleaning up the mess from

12:21

the quantum computer, Maniscalco says. So

12:24

far, it seems to work at reasonably

12:26

large scales. On top

12:28

of all that, there has been a

12:31

growing roster of achievements in Quantum

12:33

Error Correction, or QEC. Instead

12:35

of holding a qubit's worth of

12:37

information in one qubit, QEC

12:39

encodes it in the quantum states

12:41

of a set of qubits. A

12:43

noise -induced error in any one of those

12:45

is not as catastrophic as it would

12:47

be if the information were held by a

12:50

single qubit. By monitoring each

12:52

of the additional qubits, it's

12:54

possible to detect any change and

12:56

correct it before the information

12:58

becomes unusable. Implementing

13:00

QEC has long been considered

13:02

one of the essential steps on

13:04

the path to large -scale, noise

13:06

-tolerant quantum computing to machines that

13:08

can achieve all the promise

13:10

of the technology, such as the

13:13

ability to crack popular encryption

13:15

schemes. The trouble is,

13:17

QEC uses a lot of

13:19

overhead. The gold standard

13:21

error correction architecture, known as

13:23

a surface code, requires at

13:25

least 13 physical qubits to

13:28

protect a single useful logical

13:30

qubit. As you connect

13:32

logical qubits together, that number

13:34

balloons. A useful

13:36

processor might require 1 ,000

13:38

physical qubits for every logical

13:40

qubit. There are now

13:42

multiple reasons to be optimistic even

13:44

about this, however. In

13:46

July 2022, for instance, Google's

13:49

researchers published a demonstration of

13:51

a surface code in action where

13:53

performance got better, not worse,

13:55

when more qubits were connected together.

13:59

That so many noise handling techniques are

14:01

flourishing is a huge deal, especially

14:03

at a time when the notion that

14:05

we might get something useful out of

14:07

small -scale, noisy processors has turned out

14:09

to be a bust. There

14:12

have also been promising demonstrations

14:14

of theoretical alternatives to surface

14:16

codes. In August 2023, an

14:18

IBM team that included Gambetta

14:20

showed an error correction technique that

14:23

could control the errors in

14:25

a 12 qubit memory circuit using

14:27

an extra 276 qubits, a

14:29

big improvement over the thousands of

14:31

extra qubits required by surface codes.

14:34

In September, two other teams

14:36

demonstrated similar improvements with

14:39

a fault -tolerant circuit called

14:41

a CCZ gate using superconducting

14:43

circuitry and ion trap

14:45

processors. Actual error correction

14:47

is not yet happening on

14:49

commercially available quantum processors, and

14:52

is not generally implementable as

14:54

a real -time process during computations. But

14:57

Beersik sees quantum computing as

14:59

finally hitting its stride. I

15:01

think we're well on the way now, he says. I

15:04

don't see any fundamental issues at all. And

15:07

these innovations are happening

15:09

alongside general improvements in hardware

15:11

performance. meaning that there

15:13

are ever fewer baseline errors in

15:16

the functioning qubits, and an

15:18

increase in the number of qubits

15:20

on each processor, making bigger and

15:22

more useful calculations possible. Birsek

15:24

says he is starting to see

15:26

places where he might soon choose a

15:29

quantum computer over the best -performing classical

15:31

machines. Neither a classical

15:33

nor a quantum computer can

15:35

fully solve large -scale tasks, like

15:37

finding the optimal routes for a

15:39

nationwide fleet of delivery trucks. But,

15:42

Beersick points out, accessing and

15:44

running the best classical supercomputers costs

15:46

a great deal of money,

15:48

potentially more than accessing and running

15:51

a quantum computer that might

15:53

even give a slightly better solution.

15:55

Look at what high -performance computing

15:58

centers are doing on a daily

16:00

basis, says Quan Tan, CEO

16:02

and co -founder of the Finland -based

16:04

quantum computer provider IQM. They're

16:06

running power -hungry scientific calculations

16:08

that are reachable by quantum

16:10

computers that will consume much

16:13

less power. A quantum computer

16:15

doesn't have to be a better computer

16:17

than any other kind of machine to

16:19

attract paying customers, Tan says. It

16:21

just has to be comparable in performance

16:23

and cheaper to run. He

16:25

expects we'll achieve that quantum energy advantage

16:27

in the next three to five years.

16:31

A debate has long raged about

16:34

what target quantum computing researchers

16:36

should be aiming for in their

16:38

bid to compete with classical

16:40

computers. Quantum supremacy, the

16:43

goal Google has pursued, a

16:45

demonstration that a quantum computer can solve

16:47

a problem no classical computer can crack

16:49

in a reasonable amount of time, or

16:52

quantum advantage, superior performance

16:54

when it comes to a useful

16:56

problem, as IBM has preferred, or

16:59

quantum utility. IBM's

17:01

newest buzzword. The

17:03

semantics reflect differing views of

17:05

what near -term objectives are important.

17:08

In June, IBM announced that it

17:10

would begin retiring its entry -level processors

17:12

from the cloud, so that its

17:14

127 qubit Eagle processor would be

17:16

the smallest one that the company

17:18

would make available. The

17:20

move is aimed at pushing researchers

17:23

to prioritize truly useful tasks. Eagle

17:25

is a utility -scale processor,

17:28

IBM says. When correctly

17:30

handled, it can provide useful

17:32

results to problems that challenge

17:34

the best scalable classical methods. It's

17:37

a controversial claim. Many

17:39

doubt that Eagle really is

17:41

capable of outperforming suitably prepared

17:43

classical machines. But classical

17:45

computers are already struggling to keep

17:47

up with it, and IBM has

17:50

even larger systems. The

17:52

433 qubit Osprey processor,

17:54

which is also cloud accessible,

17:57

and the 1121 qubit

17:59

condor processor, which

18:01

debuted in December. Gambetta

18:03

has a simple rationale for the

18:05

way he names IBM's quantum processors. I

18:08

like birds. The

18:10

company has a new modular

18:12

design called Heron, and Flamingo is

18:14

slated to appear in 2025, with

18:17

fully quantum connections between

18:19

chips that allow the quantum

18:21

information to flow between

18:23

different processors unhindered. enabling truly

18:25

large -scale quantum computation. That

18:28

will make 2025 the first

18:30

year that quantum computing will be

18:32

provably scalable," Gambetta says. I'm

18:35

aiming for 2025 to be an

18:37

important year for demonstrating key technologies

18:39

that allow us to scale to

18:41

hundreds of thousands of qubits. IQM's

18:44

tan is astonished at the

18:46

pace of development. It's

18:49

mind -boggling how fast this field is

18:51

progressing, he says. When I was

18:53

working in this field 10 years ago,

18:55

I would never have expected to have a

18:57

10 qubit chip at this point. Now

18:59

we're talking about hundreds already, and

19:02

potentially thousands in the coming years. It's

19:05

not just IBM. Campbell

19:07

has been impressed by Google's quiet

19:09

but emphatic progress, for instance. They

19:11

operate differently, but they have

19:14

hit the milestones on their public roadmap, he

19:16

says. They seem to be

19:18

doing what they say they will

19:20

do. Other household -named companies are embracing

19:22

quantum computing, too. We're seeing

19:24

Intel using their top -line machines, the

19:26

ones they use for making chips

19:28

to make quantum devices, Tan says. Intel

19:31

is following a technology path

19:33

very different from IBM's, creating qubits

19:35

in silicon devices that the

19:38

company knows how to manufacture at

19:40

scale, with minimal noise -inducing defects.

19:43

As quantum computing hits its

19:45

stride and quantum computers begin

19:47

to process real -world data,

19:49

Technological and geographical diversity will

19:51

be important to avoid geopolitical

19:54

issues and problems with

19:56

data -sharing regulations. There

19:58

are restrictions, for instance, aimed

20:00

at maintaining national security, which will

20:02

perhaps limit the market opportunities

20:04

of multinational giants such as IBM

20:06

and Google. At the

20:08

beginning of 2022, France's

20:11

defense minister declared quantum technologies

20:13

to be of strategic interest

20:15

while announcing a new national

20:17

program of research. In

20:19

July 2023, Deutsche Telekom announced

20:21

a new partnership with

20:23

IQM for cloud -based access

20:25

to quantum computing, calling

20:27

it a way for DT customers

20:29

to access a truly sovereign quantum

20:31

environment built and managed from within

20:34

Europe. This is

20:36

not just nationalistic bluster, sovereignty

20:38

matters. DT is

20:40

leading the European Commission's development

20:43

of a quantum -based EU -wide

20:45

high -security communications infrastructure. As the

20:47

era approaches, when large -scale

20:49

quantum computers pose a serious

20:51

threat to standard encryption protocols, governments

20:54

and commercial organizations will want

20:56

to be able to test post

20:58

-quantum encryption algorithms, ones that

21:00

withstand attack by any quantum

21:03

computer, irrespective of its size,

21:05

within their own borders. Not

21:07

that this is a problem yet. Few

21:10

people think that a security -destroying

21:12

large -scale quantum processor is just around

21:14

the corner. But there is certainly

21:16

a growing belief in the field's

21:18

potential to be transformative and useful

21:20

in other ways within just a

21:22

few years. And these

21:24

days, that belief is based on real

21:26

-world achievements. At Algorithmic,

21:28

we believe in a future where

21:30

quantum utility will happen soon, but

21:33

I can trace this optimism back

21:35

to patents and publications, Menescalco says. The

21:37

only downside for her is that

21:40

not everybody has come around to

21:42

the way she has. Quantum computing

21:44

is here now, she insists, but

21:46

the old objections die hard, and

21:48

many people refuse to see it.

21:50

There is still a lot of misunderstanding. I

21:53

get very upset when I see

21:55

or hear certain conversations, she says. Sometimes

21:58

I wish I had a magic wand

22:00

that could open people's eyes. You

22:04

were listening to MIT Technology

22:07

Review, where Michael Brooks writes,

22:09

Quantum is taking on its

22:11

biggest challenge, noise. This

22:14

article was published on 4

22:16

January 2024 was

22:18

read by Michael Seitao for NOAA.