← Science In Action
Science In Action
Any more for Moore’s Law?
00:29:07
It's 60 years since Gordon Moore predicted computing power would double every two years
Loading summary
Transcript73 lines
- [00:00]
BBC Announcer
This BBC podcast is supported by ads outside the uk.
- [00:06]
Go Turkia Advertiser
The following advertisement feature is presented by Go Turkia. You're thinking about a holiday? What about a detour across the Turkish Riviera? After all, the best way to discover somewhere is to go off the beaten path. In Tarsus and Adana on Turkey's Mediterranean coast, kebab shops fight over their traditional recipes. Try the Tarsus kebab, which is thick and fatty, and the Adana kebab, very spicy. In the ancient city of Patara, the capital of what's recognized to be the world's first democratic system, discover the heritage of none other than Santa Claus. Oh, sorry, we meant Saint Nicholas, who was actually from Anatolia, not the North Pole. Patara's also home to the longest beach in Turkey, where Careta Careta turtles come out to play. Experience even more history along the St. Paul Trail through mountains and forests. Follow the same route the saint took on his first missionary journey. Nowadays, it's also a trail for the energy gel lovers. A bike tour around here is definitely set to work up a sweat. So there you have it. Turkey is med flavor history, St. Paul and St. Nick. Time to start packing.
- [01:22]
BBC Announcer
At the BBC we go further so you see clearer With a subscription to BBC.com, you get unlimited articles and videos, hundreds of ad free podcasts and the BBC News Channel streaming live 24. 7 from less than a dollar a week for your first year. Read, watch and listen to trusted independent journalism and storytelling. It all starts with a subscription to BBC.com find out more@BBC.com unlimited.
- [01:54]
Gordon Moore
What has become known as Moore's Law was something that came out of an article I wrote in 1965 Electronics magazine. This was at a time when we were really getting the first generations of integrated circuits to market. The first One really was 1961, before there was any of the markets. This was four years after the first ones were out, and they had increased slowly in complexity. Or it increased in complexity. Maybe it wasn't slowly.
- [02:23]
Ronan Pease
I met Gordon Moore at Intel HQ in Silicon Valley in 1997, just ahead of the 50th anniversary of the invention of the transistor, the little electronic switch that is at the heart of basically all electronics and all computing. It was also a couple of years after the 30th anniversary of his observation about progress in electronics, which later became known as Moore's Law.
- [02:47]
Gordon Moore
I was given the job of predicting what would happen in the component industry over the next 10 years, and I looked at what we had done in the preceding few years and saw that the number of components as transistors and resistors we were putting on an integrated circuit had about doubled every year. So I took the rather bold and frankly naive idea and said, okay, we're going to do this for another 10 years. And at the end of that time, Instead of putting 60 components on a chip, we'll be putting 60,000 components on a chip. And that'll be the cheapest way to buy a piece of electronics.
- [03:25]
Ronan Pease
I'm Roland Pease and I've made my trek to Silicon Valley for the BBC to learn about the invention of the transistor and the people who made it central to our lives and to hear about the challenges of keeping that progress growing.
- [03:37]
Gordon Moore
I have been really amazed that we've been able to stay on the trend we've been on as long as we have. But in real respect it's become a self fulfilling prophecy. The industry understands the rate at which we're making progress and increasing complexity. The companies know if they fall behind that trend line, they lose out. And if they can get ahead of that trend line, they can get a proprietary position. So in real respect, Moore's Law has gone from recording what's going to happen to driving what is happening in the industry. Now this gets tougher and tougher because in order to keep doing this, we have to continue to make things smaller and smaller. That's what gives us this tremendous decrease in the cost of electronics.
- [04:24]
Ronan Pease
This week, Moore's Law turns 60. And for this edition of Science in Action, I'm marveling at how an extrapolation based on the first baby steps of microelectronics has kept going and going, in essence, if not in detail, for six decades. As you said, those 1965 chips had 60 components on them. In 1997, when I met him, the Pentium II had 7.5 million components. Today a high end consumer chip might have 10 billion or more for the most part. By making them smaller and smaller, as Gordon Moore said, and that despite the fact that the industry always was worried the good times might end, the atomic.
- [05:08]
Gordon Moore
Nature of matter starts to impact what the electrical properties of the devices look like. Something between a tenth and a hundredth of a micrometer minimum feature size, I believe we're going to find that the physics changes sufficiently. Devices don't work the way let you make good circuits anymore. And that's going to be a real limit in how far we can go just making things smaller. Now if we stay on the curve, we've been on the so called Moore's Law curve that happens something between 2010 and 2020. In fact, one of my colleagues says it happens on July 17, 2017, but.
- [05:50]
Ronan Pease
I didn't notice anything in 2017.
- [05:52]
Stan Williams
Did you?
- [05:53]
Ronan Pease
There are two fundamental challenges to keeping Moore's Law going in its pure form. The physics, like Gordon Moore said of the components being smaller, getting the voltages and currents to do their things reliably with fewer and fewer atoms in the active parts of the device, and, and the manufacturing of the circuits with millions or billions of components each many thousand times finer than the width of a human hair. But the basic way of doing this, using complex optics to project the circuit design down onto treated silicon photolithography, remains the same. It's just they've got much, much cleverer at focusing the light to do the fine detail.
- [06:34]
Sri Samavedam
IMEC is a non profit R and D organization and we work with the entire semiconductor ecosystem. Material suppliers, equipment suppliers, the companies that run the fabs, the memory makers, as well as idms like Intel, Samsung, even the system companies like Apple, Qualcomm, amd, et cetera. So what we try to do is we try to anticipate the challenges in the roadmap and then we try to identify which are the most manufacturable paths forward.
- [07:08]
Ronan Pease
Based in Belgium, IMEC is one of the developers that the big fabrication plants, the fabs turn to to keep Moore's Law going. Sri Samovedam is their vice president for cmos. That's silicon technologies in general.
- [07:21]
Sri Samavedam
You know, what Moore's Law represents is, in my view, it's the confidence in the collective ability of smart people to innovate continuously and solve difficult technical problems. And today we are able to pattern features of the order of several nanometers.
- [07:40]
Gareth Nelson Davis
So you mentioned there you're getting down to a few nanometers. When I met Gordon Moore, he was talking about things and in fact the Pentium chip there had transistor features. So the details on these circuits were around 330 nanometers. And they were worried that it was.
- [07:59]
Ronan Pease
All going to stop.
- [08:00]
Gareth Nelson Davis
I mean, now you're down to a few nanometers, so you're 100 times finer. And that presumably also means you're getting tens of hundreds or hundreds of hundreds more components into the circuitry. The challenge is to do it without ever increasing the cost of the process in a way.
- [08:17]
Sri Samavedam
Right. So these days what we do is it's becoming more and more difficult to scale the dimensions. So we're stacking chips on top of each other. We're also putting the chips side by side in a chip package. So these days we track transistors in packages. You know, you can find about, you know, 200 to 300 billion transistors in a package. And we expect that in the 2030 time frame, it'll reach over a trillion transistors.
- [08:45]
Gareth Nelson Davis
In the process of photo lithography, it's basically a projection system. Instead of making a small picture on a big screen to watch at a cinema, you're projecting backwards the pattern of these components onto tiny little silicon wafers, but by the hundreds at a time.
- [09:06]
Sri Samavedam
That is correct. So the wavelength of light determines how small of a feature you can print. And so over the years, we have scaled the wavelength of light that is used in this photolithography process. We used to start with a 365 nanometer line width, which is ultraviolet, I guess. Yeah, it is ultraviolet. And now we are in extreme ultraviolet regime, where we have wavelength of about 13 and a half nanometers.
- [09:39]
Ronan Pease
13 and a half nanometers, yes.
- [09:41]
Gareth Nelson Davis
I mean, this is getting towards the x rays. I mean, it's interesting because it's one of the techniques which they were talking about at intel in 1997.
- [09:50]
Sri Samavedam
Yeah, you know, it is fantastic.
- [09:51]
Gareth Nelson Davis
And the optics, I mean, you say, oh, well, we'll just do it with extreme ultraviolet. But the demands that puts on your manufacturing process are pretty amazing.
- [10:03]
Sri Samavedam
Yeah. The UV lithography tool is an engineering marvel. It involves shooting laser light on droplets of metal to produce these UV rays. The light is projected onto the wafer using extremely precise lenses. It is quite fascinating that it actually.
- [10:19]
Gareth Nelson Davis
Works for me, that's what's so extraordinary is you engineers seem to be able to find an extra 5% here of 5% there. And the constraint is this all has to be done. It only works if you can make these chips. I mean, I know you make them on these big 30 centimeter disks, but they have to be processed incredibly quick because that's the only way it can be cheap enough to sell.
- [10:43]
Sri Samavedam
That is right. So the throughput is important. Building a semiconductor fab is quite expensive, you know, tens of billions of dollars. And if you are not able to utilize the tools to the fullest extent, it's hard to justify the economic investment. So it's a highly automated and optimized process. Just to give you an example, a chip that is made at the most advanced node may consist of hundreds of different process steps. And it may take anywhere from two months to four months, from start to finish, to come out of the fab. And this is in a factory that is running 7 by 24.
- [11:22]
Gareth Nelson Davis
And in a way, people, the companies only keep ahead of their competition by, you know, having a few months head start on some process. I mean, what year are you looking at at the moment? We're now seeing things with features of 5 or 3 or 2nm across. But what are you thinking about for? I don't know, do you look ahead 10 years or what?
- [11:44]
Sri Samavedam
Yeah, so we look ahead anywhere from five to 10 years. Right now we are working on a 7 Angstrom technology. An Angstrom is a tenth of a nanometer, which we expect will go into production in the 2032-35 time frame. That's our focus right now. The 2 nanometer technology node is in production starting this year. After the 2 nanometer comes the 14 Angstrom node, the 10 Angstrom node, and then the 7 Angstrom node.
- [12:15]
Ronan Pease
3 Summerve Adam of IMEC. You might not be surprised to know there's another version of Moore's Law, sometimes called Rock's Law.
- [12:22]
Gareth Nelson Davis
But.
- [12:22]
Ronan Pease
But the cost of building the next fab also keeps rising, though not quite as fast, which is why the economics still work. A reminder, this is science in action from the BBC. Wondering how much more we will get of Moore's Law.
- [12:39]
BBC Announcer
At the BBC, we go further so you see clearer. With a subscription to BBC.com, you get unlimited articles and videos, hundreds of ad free podcasts and the BBC News Channel streaming live 24. 7 from less than a dollar a week for your first year. Read, watch and listen to trusted independent journalism and storytelling. It all starts with a subscription to BBC.com find out more@BBC.com unlimited.
- [13:11]
Ronan Pease
Of course, all the while, there have been other technologies knocking at the door, saying they're ready to take over when Moore's Law gives out. One of them, talked about breathlessly these days, is quantum computing the solution to all our ills? If you listen to some commentators, the paper that kicked off the discussion of those possibilities was published in 1992, when Moore's Law was not yet 30 years old. They're still working on the best way to build a quantum machine. Meanwhile, the fastest computers today run 10 million times faster than when that paper came out. It's not to denigrate quantum computing, but to underline the extraordinary power of the incremental improvements in the old technology. Computer scientist Scott Aronson of the University of Texas would love more realism in the way that we talk about quantum computing.
- [14:03]
Scott Aaronson
Well, quantum computing in the early 1990s had barely even reached the Charles Babbage stage. And I would say that today, that it's about at the stage that classical computing was at in the 1940s. You know, we have devices, they are programmable, they have typically 50 to 100 qubits. We can do thousands of operations on them. You know, we can do some very contrived tasks that plausibly a classical computer would not have been able to do within the lifetime of the universe, which is quite interesting.
- [14:38]
Gareth Nelson Davis
They're sort of demonstrations. I mean, they're doing the things which quantum computers can do. Well, to demonstrate, I guess.
- [14:43]
Stan Williams
Yeah.
- [14:44]
Scott Aaronson
Now, the harder job that we have is that it's not enough for us to just demonstrate something. We have to beat classical computing, which already exists, or else it's not useful.
- [14:55]
Gareth Nelson Davis
Absolutely. In the 30 years since these proposals were first becoming solid, in the meantime, Moore's law has meant that the classical computers are now billing.
- [15:08]
Scott Aaronson
It's meant that the bar keeps getting higher and higher. Yes. And not only that, but people keep coming up with better and better algorithms for the classical computers. And generative AI has started to work, which means that often, even when we don't know the algorithm, AI can just sort of guess it. Right. And so that is all set a much higher bar that quantum computing has to beat in order to be useful.
- [15:35]
Gareth Nelson Davis
I mean, is there within quantum computing something like Amore's Law with the number of qubits, or the complexity of calculation or procedure improving in some kind of exponential way with all so many people working on it?
- [15:51]
Scott Aaronson
People have talked about that, but I think it's very early to say. I mean, these state of the art devices that have 50 to 100 qubits.
- [15:58]
Gareth Nelson Davis
Right.
- [15:58]
Scott Aaronson
It's very hard to pass an exponential through numbers that end at 100. I'd like to see them get up to, you know, 10,000 or 100,000 first.
- [16:08]
Gareth Nelson Davis
Ah, yeah. Is there a thing with quantum computing? And I think you're sort of hinting at this that there's a threshold involved. I mean, you know, in, back in, back in the 1960s, a small computer using a few transistors could help you land on the moon. And so a small computer is very helpful. Is there, is there a point where suddenly the richness of what quantum computing might offer will become apparent?
- [16:37]
Scott Aaronson
There's one very, very important threshold which is called the error correction threshold. Right. So as the components get better and better, as we get better control over them, there will come a point when we can use error correcting codes in order to simulate effectively perfect components. Okay. And that's been the real engineering goal of this field for the past 30 years. And just within the last year, we have seen arguably the first demonstrations of an error corrected qubit. So that's a very key threshold that, you know, once you've crossed it, then you ought to be able to scale up to as many qubits as you want and as many operations on those qubits as you want. We're not there yet, but we are a lot closer than we were when you were reading about this in the 90s.
- [17:29]
Ronan Pease
Scott Aronson. Another technology that's endlessly been in the wings is optical computing. Electrons have all sorts of unique selling points when it comes to circuitry, not least the way they interact with voltages, giving the engineers the control they need to run a program. But there are disadvantages, not least that currents generate heat, a limitation Gordon Moore highlighted in his 1965 prediction. And with today's circuits, currents introduce delays that make it hard to synchronize components across a chip. Two papers just published in Nature show how light can massively boost the performance of otherwise conventional chips. One came from California Startup Light Matter CEO Nick Harris.
- [18:15]
Nick Harris
@ this point, computer chips are doubling in package size every 18 months instead of keeping the same size. And the power is going up exponentially. It's an unsustainable trend. And so what we're doing is we're leveraging light, which follows a completely different set of physics, to find a new scaling pathway to make computers better, more efficient. And some of the properties that are really interesting there are the fact that you can use multiple wavelengths of light to perform calculations in parallel. Light is also very fast. The carrier frequency is 193 terahertz. And so you have an incredible amount of bandwidth. You can have clock rates of 50 gigahertz and beyond. So those are a couple of the exciting things.
- [18:58]
Gareth Nelson Davis
So in a sense, the high bandwidth that you're talking about, multiple colors, so going from one component to another, that's a bit like we use fiber optics to transmit the Internet and so on across the Atlantic. With fiber optics, that's a very efficient way of getting information. Point from point A to point B.
- [19:19]
Nick Harris
Yeah, that's right at the end of Moore's law and Dennard scaling, which is where we're at. There are only two vectors for making progress in computing. One of them is building bigger chip packages. If you'd like to double the performance of your processor, you double the amount of silicon area per package. The other technique is that you link together lots of chips and those links are transitioning to be photonic now. And I think by 2027 you'll see a lot of these AI supercomputers are going to be based on photonic links. And what they're going to enable is these giant million GPU clusters that act as a single supercomputer. And that's something that's not possible today. Today you have very small groups of chips that are communicating very quickly and they're loosely connected with low bandwidth connections. But that's all about to change.
- [20:12]
Ronan Pease
The other thing you mentioned though, and.
- [20:13]
Gareth Nelson Davis
This is one which has always intrigued me, you said that you could use light also to do the processing.
- [20:19]
Nick Harris
Yeah, that's right. So what we showed in this publication was that you could do the mathematics that lies behind deep learning and artificial intelligence, which is basically multiplies additions and these sorts of math using light and the way you do it using what's known as an optical modulator. And imagine you have a garden hose and you're squeezing the garden hose and you're looking at the water coming out the other end. You're encoding that squeezing into the stream of water that's going through sort of.
- [20:51]
Gareth Nelson Davis
Pulses of water coming out as you do that.
- [20:53]
Nick Harris
Exactly. So you're modulating the intensity of the light in this analogy. And we think that over the coming decade it could be the case that this technology can be deployed in data centers and used to accelerate AI workloads and hopefully drive down the energy consumption and drive up the performance per area, which is a critical metric that Moore's Law has enabled over the past 60 years.
- [21:20]
Gareth Nelson Davis
One reason I'm doing this is, you know, I heard from Gordon Moore himself about 30 years ago, but from lots of other people, that things were getting too hard with conventional electronics and different approaches were going to be taken. But it's this juggernaut that silicon and transistors and so on has always won through. And all the clever guys lost out. It seems though you keep trying. Are you worried that you could put in this effort and nevertheless you will be sort of bypassed by the traditional methods?
- [21:54]
Nick Harris
You know what's so cool about the way that we've approached this is we're using cmos. Our devices live next to transistors within microns. And so we're leveraging all of the multi trillion dollar infrastructure of the semiconductor industry. And I think when you look at alternative computing approaches of the past, they really haven't had that opportunity. So we get to leverage that entire economy of scale, the supply chain and everything else to make this possible. And indeed, if you look at photonics. It's already starting to be integrated with computer chips, and that's in the form of interconnect. We do that at Light Matter, but there's also been announcements at Nvidia, Broadcom, and Cisco. And so this is starting to be mainstream in a way. The technology set using it for computation is another step forward, but it's already kind of in there.
- [22:44]
Ronan Pease
Nick Harris of Light matter. Not quite 20 years ago, we featured on Science in Action the discovery of a new component that promised to displace the transistor as the tiny electrical switch at the heart of computing called the memristor. A few years later, I made a whole edition of Discovery on the promise it would bring new life to Moore's Law. I feel I'm still waiting. So I called Stan Williams then reshaping the future at Hewlett Packard, now still working on it at Texas A and M University for an update and a reminder of what makes a memristor so different.
- [23:21]
Stan Williams
A memristor is an electronic device that remembers its history, which is where the word mem comes in. And what that means is that you can essentially program it in an analog fashion. In other words, instead of digital pulses through the device, you apply a constant current or voltage, for instance, and it will change its resistance in a uniform fashion. That actually is a very large advantage right now, because the digital world is running out of material and energy. Whereas memristors have this capability of storing information in an analog fashion, such that one small device can store a very large amount of data, for instance, as a real number.
- [24:22]
Gareth Nelson Davis
So in other words, you're getting more computation, as it were, per square nanometer of material.
- [24:29]
Stan Williams
That is correct. And in fact, the magic of Moore's Law, or as I call it, Moore's scaling, was that industry was able to, for a very long time, double the number of transistors on a circuit every 18 months or so. So with that exponential increase, technology advanced further and more rapidly than ever in human history. With memristors, you actually have the possibility for a similar type of exponential advantage. And that is as you are able to control the analog states better and better, the higher the resolution or the higher the accuracy that you can have in your device. That means that you can scale your computation with time just as the. As the surrounding circuits get better. And so there's actually no need to add more devices on a circuit. There's no need to use more energy. You can essentially perform more and more computation using the same footprint and the same Amount of energy.
- [25:39]
Gareth Nelson Davis
I have to say, the way you describe it sounds quite complicated. I'm thinking that a transistor is a bit like a water pipe with a. With a tap on the side that you can turn on and off, and that's either flowing or not flowing. It sounds like you're more like traffic flow with these memristors or a grayscale that's darker or brighter, according to, I don't know, the current or voltage you're applying.
- [26:06]
Stan Williams
Yes. And that's, you know, really the difference between the analog and the digital worlds. In the digital world, you know, it's either black or white. Your switch is more like a wall switch. You turn it off, you turn it on back and forth. Or you have a screen that can only display black and white pixels, whereas in the analog world, you have the entire range of gray. And so what we're trying to do is add more and more levels of gray. There are actually quite a few startup companies right now that are making simple products that are intended for very specialized purposes. And that's, of course, the way transistors got started in the first place. The first product for a transistor was a hearing aid. So it was a very specialized application. And it grew in and eventually took over from vacuum tubes because they had all of these advantages in terms of size and power requirements. Well, you know, something similar, I think, can very well happen in the digital to analog transition. We've been in a digital world for a very long time for very good reasons, but our circuitry that we're building right now is getting better and better in terms of its analog precision. And so what I'm viewing is a type of Moore's Law, in which we add one equivalent bit of precision to our analog states every year or two. And so what that has is the potential for doubling the amount of computation that can be performed, any amount of data that can be stored every year, and there's really no reason that can't go on for a couple of decades. So that's a lesson from Moore's Law is find a technology that scales. Find something that you can continuously improve, and then ride it. That's the new horse that we're trying to get on.
- [28:06]
Gareth Nelson Davis
One of the phrases I associate with memristors is this word neuromorphic computing, which I take to be a bit like the brain. And I see a lot of papers saying that it would be good for artificial intelligence.
- [28:21]
Stan Williams
Yes. And that's in part because memristors do strongly resemble synapses in brains in terms of their properties. And so people are building machines, or chips, if you will, that are trying to mimic at least particular processes that occur in human brains and do it with a very low amount of energy.
- [28:47]
Gareth Nelson Davis
These devices do sound very fancy. Do you imagine in, I don't know, 10, maybe 15, 20 years time, laptops and our mobile phones will have these kinds of devices as pretty standard?
- [28:59]
Stan Williams
I think it's going to be a lot sooner than that. There are many foundries that are already using them and building them. And in fact, you know, a lot of the work is, I'll say it's kind of secretive. There are circuits that I know about that contain memristors within them, but that's not advertised. I mean people, you know, whether it's because of proprietary issues or secrecy or fear of intellectual property or whatever, people don't talk about it or, you know, companies don't talk about, you know, that they're actually using them today for storing firmware in a chip, for example. So there's, there's, there's a lot of, there actually are a lot of chips out there already. I would, I would say that, you know, to my knowledge it's probably on the order of hundreds of millions of chips already exist with some small number. So, and what that does is it is it gives the foundries practice and understanding about how to make the devices, how to, how to handle the materials, how to improve. And so it's, you know, it's a slow build up process. I think what a lot of people forget is how long it took from the first transistor in 1947 to the first processor circuit. That was a period of something like 16 or 17 years. So progress appears to be slow, even though it's actually exponential all along. But then you hit a point in the trajectory where now as you start doubling, it has a major impact on technology. So I would say that we're just, we're just shy of that inflection point right now.
- [30:39]
Ronan Pease
Stan Williams, master of the Memrista revealing secrets I didn't know. Six decades of uninterrupted progress really is something to celebrate. It's not as if cars or planes are billions of times better than they were in the 1960s, nor drugs routinely that much more potent. Which is why I thought it was worth devoting an edition of BBC Science in Action to this anniversary. I'm Ronan Pease. The producers are Alex Mansfield and Gareth Nelson Davis. And this one more time is Gordon Moore.
- [31:10]
Gordon Moore
I had no idea that that was going to prove to be very predictive over that decade and one of my friends at a California university I think named it Moore's Law. Since then the concept of Moore's Law has been extended to a lot of things in the industry. Kind of anything that changes geometrically is called Moore's Law. And I'm perfectly willing to take credit credit for all of it.
- [31:38]
Go Turkia Advertiser
The following advertisement feature is presented by Go Turkia Ever thought about taking a detour? Turkey is known for its breathtaking beaches and mouth watering kebabs. But did you know it's also a hub of ancient histories, a verdant land for produce and wine, and home to Michelin star cuisine that surprises and delights? What more could you ask for? Look no further than the Aegean region of Turkey. It has it all. Discover culinary excellence in Erla Izmir by sampling local produce in restaurants along the way. Experience history in Teos, an ancient Ionian city in Sephora, where they loved a good glass of wine so much they built temples in honor of Dionysus, the God of wine and theater himself. And feel the sand beneath your feet along the Aegean's stunning beaches. Or maybe the wind through your hair kite surfing along the Elecita coastline. Turkia isn't just about the sun lounger, it's about the adrenaline. If you thought you had Turkey all figured out, try out the Turkogean lifestyle.
- [32:48]
BBC Announcer
At the BBC. We go further so you see clearer. Through frontline reporting, global stories and local insights, we bring you closer to the world's news as it happens. And it starts with a subscription to BBC.com giving you unlimited articles and videos, ad free podcasts and the BBC News Channel streaming live 24. 7. Subscribe to trusted independent journalism from the BBC. Find out more@BBC.com join.