For most of my life, I’ve not slept well. When I was 10, I’d stay up all night reading books under my duvet, using my alarm clock’s bright blue display as a reading light. For each of the past 3 years, I’ve had exactly 6 hours 51 minutes of sleep on average.

I’m now sleeping more than 7 hours on a consistent basis. I’ve invested a LOT of time and effort into improving my sleep. I no longer get tired in the afternoon, and I DO get tired in the evenings when I want to go to sleep. It’s awesome.

Making sleep a priority is essential. I prioritise sleeping well above almost everything else.

This blog post provides an instruction manual for improving your sleep, based on everything I’ve learnt. I’ve tried most things, and this is what actually worked for me.

Start with the basics, then go up the levels if you still need them.

Let me know how it goes for you!

Quick summary

• Decaf: stop drinking caffeine by 2pm

• Go dark: avoid bright lights in the evening by using red lens glasses and red lamps

• No screens: keep your phone and laptop away from your bedroom

• Consistency: stick to a regular wakeup time

• Cool room: use a light duvet and/or reduce your bedroom’s temperature by opening the window before bed

Level 1: Cheap Fixes

• Have a consistent wake-up time

  • What time do you want to wake up at? For me, it’s 8am. This gives me enough time in the morning to do my morning routine, exercise, and commute to work in time for our 10am standup.

  • I try very hard to not change my wake-up time. Even if I go to bed late or can’t sleep, I will still wake up at 8am.

  • If you start moving your wake-up time because of late nights, you’ll fall into a spiral of being less tired the next evening, then not being able to fall asleep, so you move your wake-up time even later.

  • Your wake-up time is the anchor for your whole sleep routine. I start my evening routine at 11pm, 9 hours before my wake-up time. This gives me enough time to switch off in the evening from work.

• Bright lights in the morning

  • Once you wake up, open your curtains and bask in the beautiful morning sunshine.

  • Consider stepping outside and getting some fresh air.

  • Bright light first thing in the morning will tell your body it’s daytime!

• Drink less coffee

  • If you don’t sleep well the night before and you’re super tired in the afternoon, don’t solve it by drinking 5 coffees. Remember that you will soon become your future self: be kind to them.

  • Get a decaf coffee, swap coffee for tea, include some herbal tea in there (rooibos and jasmine are lovely!), do whatever works so you consume less caffeine.

• Avoid alcohol

  • Alcohol might help you fall asleep, but it’s terrible for your sleep quality.

  • My watch gives me a “sleep score”, and even after just 1-2 drinks, my sleep score drops by 20-30 points.

  • If you’re serious about improving your sleep, drink less alcohol, or cut it out entirely.

• No late-night snacks or drinks

  • I try to avoid any food or drink after 7pm. I don’t always succeed at this, but I do most of the time.

  • You need to give plenty of time for your digestive system to process your food before going to bed, and you don’t want to wake up in the middle of the night to pee.

• No phone in the bedroom

  • Smartphones are the enemy of sleep. You’re probably addicted to your phone. So am I. It’s not ideal.

  • I use AppBlock to make my iPhone unusable in the evening, and I keep my phone in a different part of the house. I never bring it into my bedroom in the evenings.

  • Buy a standalone alarm clock to get you out of bed. Put it on the other side of your room so you have to get out of bed to switch it off. Resist the urge to go back to bed.

• Red lens glasses

  • Cavemen didn’t have bright screens and LED lamps.

  • We inundate ourselves with bright lights all day, extending late into the evenings, which ruins our circadian rhythm.

  • I work late, and my work involves staring at a bright computer monitor. I used to think my most productive period was from 10pm-2am, because I felt wide awake.

  • Buying red lens glasses was the single most impactful thing I’ve done for my sleep.

  • I put them on around 9pm, and for the first time in years, I feel tired before going to bed. It’s amazing.

  • They block out blue and green lights, and you can get them for £23 from Amazon, or you can get them on prescription here.

• Keep your room cool

  • An hour before going to bed, open your window to let cool, fresh air into your room.

  • Your body needs to drop in temperature to fall asleep properly. Aim for 16-19°C.

• Eye mask

  • If you don’t have excellent blackout blinds yet, buy an eye mask.

  • Cheap eye masks rest on your eyelids and feel uncomfortable. Instead, pay for a good one, e.g. Manta (£29) or MyHalos (£10).

• Foam ear plugs

  • Back in the day when we needed to be super vigilant at night in case a bear ransacked our mud huts, it made sense for us to be light sleepers. A bear hasn’t ransacked my mud hut for a long while, so I want to dull myself to nighttime noises.

  • There are feisty foxes in my garden that fight during the night, and parakeets that sing at an ungodly hour. Damn them!

  • Solve this by wearing ear plugs. You can get cheap foam ear plugs on Amazon - try out a bunch of different ones to see what works for you.

Level 2: Advanced Sleeper

• Red light bulbs

  • I’m writing this at 10pm. Walking past my house right now, you’d see a room emanating a spicy red glow.

  • All the bulbs in my bedroom and office are RGBW smart bulbs. At a set hour, they all go red, and sleep music starts playing out of my speakers. It’s a clear signal to my brain that it’s sleep time.

• Darker screens

  • I like working, and I work late. To reduce the brightness of my screen, I use a chrome extension called “Dark Reader” in the evenings.

  • It turns every website into “dark mode”, even if the website doesn’t have a native dark mode.

  • This is a nice complement to wearing red lens glasses and having red lights on in the room.

• Journaling

  • Being a human can be pretty stressful. We make mistakes all the time, we live in a crazy chaotic world, nothing makes sense.

  • I’ve found that reviewing the day, describing what I did, writing down how I’m feeling, and doing a “brain dump” into my journal has helped me reduce how much I ruminate in bed.

  • I journal in Airtable, but you could do this in any random app, a google doc, a notebook, or wherever works for you. Don’t stress about the format - just do it.

• Meditation

  • Ommmmmm.

  • Meditation is another classic technique for helping you notice what your mind is doing, calm it down a bit, and start to slow down your thoughts in time for going to bed.

  • You could get a fancy meditation app, or you can just type in 5 minute meditation into Spotify. Again, just do what works - don’t stress about the format.

• Breathe through your nose

  • There’s a reason why “mouth breather” is an insult.

  • Breathing through your nose has lots of health benefits. Breathing through your mouth can also make you snore. Bad for your partner’s sleep!

  • It’s a bit weird, but you can train yourself to breathe through your nose by taping your mouth shut with surgical tape. I did this before going to sleep for a few weeks, and now I breathe through my nose all the time (including while running! Which is a great feeling too).

• Custom fitted ear plugs

  • Foam ear plugs are fine, but you know what’s better? An ear plug that is moulded to the exact shape of your ear canal.

  • If you’re in London, check out this or this. You have to go to a specific place, where someone will squeeze a bunch of gel into your ear to make the mould. It’s a weird experience.

  • I wear these 90% of the time. Sometimes I swap back to foam ear plugs for a night for variety, but the custom ear plugs are much more comfortable. I barely notice I’m wearing them, they don’t fall out, and they block out most noise.

Level 3: Sleeper Pro Max

• Breathe clean air

  • 200 years ago, we started this thing called the “Industrial Revolution” and it filled our lungs with soot.

  • If you live in a city, the air quality is probably bad.

  • I have two Xiaomi air purifiers at home and they’re great.

• Blackout blinds

  • Eye masks are really annoying and they sometimes fall off your face. Then the blinding light from the sun that will one day engulf the earth wakes you up. I’m trying to sleep!

  • Upgrade your see-through curtains to a set of fitted blackout blinds. These block out 100% of light. It’s magic. It’s also a fun party trick to show your friends how dark your room is. I’m really cool.

  • I bought the non-smart versions of these - find whatever works for you.

  • You can make them “smart” in a DIY way by getting a blind motor. That’s another really super duper cool party trick - “Hey google, open the blind.”

  • And of course, make them open automatically in the morning and close automatically at night!

• Buy a good mattress, pillow, and bedding

  • I don’t know how to buy a good mattress. It’s a terrible consumer experience: you can’t really test it beforehand; even going into a mattress shop and lying on a few of them doesn’t give you much information about which you’d sleep best on.

  • I read lots of reviews and bought this mattress, this pillow, and these bedding. Are they the best? Who knows. I like them.

• Seasonal duvets

  • I used to have only one duvet. This is a mistake, especially if you sleep hot.

  • I now have a 10.5 tog duvet for the winter, and a 4.5 tog duvet for the summer.

• White noise machine

  • Ear plugs are great, but they don’t block out all sound. To dull out any other external noises, I use this white noise machine.

  • It goes off and on automatically using a Shelly Plus Plug. I taped the light on the plug using blackout tape.

• Temperature-controlled bed

  • This is perhaps the most fancy thing in this entire blog.

  • During the Summer, even with a light duvet, I often struggle to fall asleep due to being too hot. This improved a lot once I got an Eight Sleep.

  • It’s super expensive, but for me it’s been really good. If you don’t have an AC in your room and you sleep hot, I recommend it.

• Portable AC unit

  • The Eight Sleep works great for 95% of hot days, but if the room gets too hot during a heat wave (>30°C), it’s not enough.

  • In these instances (which is ~1 week in a year in the UK), I use a portable AC unit. I bought a Black + Decker one and it’s great.

Conclusion

Sleep is super important. You’ve read this far — you know it’s important. Improving your sleep takes many small experiments over a long period of time. Don’t rush to do everything. Do the cheap and easy things first, and consider doing the more time-intensive and expensive things later if you’re still struggling to sleep well.

You can and will sleep better. I’m rooting for you!

Mornings are a great opportunity to build habits that will improve your life, like meditation, exercise, drinking plenty of water, and eating a healthy breakfast. Once the workday’s started, it’s hard to carve out the time to look after yourself, but the morning provides a few sacred hours that are yours to design.

Every morning routine I’ve designed until now worked while my willpower was strong and conditions were perfect. But willpower is a tenuous thing, and life is all about surprises. At some point, I’d wake up late, or open my phone and doom scroll for a few hours, or feel too tired to follow the routine. I’d tell myself that I just couldn’t hack mornings — ”I’m not a morning person!” — and I’d throw the routine out the window.

When I designed a morning routine, I’d spend an hour thinking it through and writing it down, then I’d hope that it would work for every morning thereafter. If I failed to follow the routine on a given morning, I wouldn’t know if it was due to me being lazy and disorganised, or if the routine itself wasn’t suitable. The routine was static, and I hadn’t set myself up for rapid iteration and improvement. If my sleep schedule changed, or I had new commitments in the morning, or if anything else changed from the ideal conditions I imagined when designing the morning routine, I expected my cognitively impaired morning self to improvise and design a new routine on the fly. As expected, I would fail, and I’d feel rubbish and angry with myself.

This all changed last month, due to A6 ruled index cards and a pen.

How did I design an improving-and-robust-by-default morning routine?

I know that I will sometimes fail to follow my morning routine. I also know that every morning is different, and there isn’t a one-sized-fits-all morning routine. I needed a system that helps me adapt to changing circumstances, and has something simple and reliable at its core that is easy for me to pick back up if things don’t work out on a given day.

Solve tomorrow morning first

Instead of asking myself “what morning routine do I want to follow over the next year?”, I now ask myself “what do I want to do tomorrow morning?”, and I make a simple checklist on a single A6 ruled index card.

Imagine the morning: I open my eyes, feeling groggy and tired. My alarms are blaring at me, so I get up and turn them off. I walk over to the table and see a piece of paper. I read it, and it has simple instructions to follow. My brain isn’t working yet, so it’s nice to be told what to do. These are simple tasks, I can do this. I’ll drink some water from the full water bottle that some nice gentleman has left for me on the table, then I’ll brush my teeth. Two ticks. Meditate? Hmm, I’m not sure I can be bothered to do that. But if I don’t, I can’t tick this box. I want to tick all the boxes, so I guess I will meditate. I’m now being told to put on some running clothes, and leave the house to go running. I feel energised to run after the meditation, and my running clothes have appeared like magic next to the piece of paper, so off I go!

Instead of trying to solve a big scary problem like “what’s the best morning routine?”, I’m now focused on answering every day the simple question of “what do I want to do tomorrow morning, and how can I make my morning self’s life as easy as possible?”.

Make improvements every day

My previous morning routines were designed by the most intentional and ambitious version of me, when I felt most excited about improving my wellbeing and life outcomes. After creating the routine, I’d cross my fingers and hope that lethargic morning me would follow it. This didn’t work. These two versions of me weren’t talking to each other. My morning self was frustrated that the routine didn’t factor in X or Y new circumstances, and my intentional self was angry that my morning self was too lazy or disorganised to do what I wanted. They hadn’t been to couples counselling yet, so nobody had prompted them to talk to each other about their feelings.

My new system is reinvented every day, and my intentional self is in an ongoing conversation with my morning self. Every evening, I review the card from that morning. This is an opportunity to identify weaknesses or flaws in my plan. For example, perhaps my checklist said “go running” and then “meditate”, but after running, I wanted to jump in the shower and head straight to work. I didn’t want to pause for a meditation, so I didn’t meditate. I note this down on the card. In the evening, I think to myself: “what about meditating before running? I’ll try that tomorrow!”. I’m refining and improving the plan every day, learning what works and what doesn’t, enjoying this new playground for experimentation.

Every morning is different too. On some mornings, I run. On some mornings, I workout. On some mornings, I commute to work. Every time I create a checklist for the next morning, I factor this into the plan. I try to set myself up for success, given the conditions I will find myself in.

Morning routines start the night before

At 10:40pm every evening, my phone starts buzzing telling me to set intentions for the next morning. At the same time, my bedroom lights go red, and sleep music starts playing from my speakers. I walk to my table and create a checklist for the morning.

In order to succeed at my morning routine, I must first succeed at my evening routine. I know that my energy levels are high in the evenings, so it’s been quite easy for me to establish the habit of setting morning intentions. It’s easy, fun, and only takes a few minutes. Even if I fail to follow my morning routine on a given day, my evening routine still stands strong, and I’m able to give it another go the next day. I think this could apply to any solid routine someone has, e.g. having it be the default action you take after brushing your teeth in the evening, or something you do before having lunch, or whenever you’re confident you can set your morning intentions reliably.

However, if I don’t get enough sleep, or if my sleep schedule is all over the place, even the best intentions won’t be enough to convince morning me to follow my instructions. I try to sleep at the same time each day (even on weekends), and every time I have a bad night’s sleep, I diagnose why, and I prioritise fixing it. Common reasons for me not sleeping well include light coming in to my room hours before my intended wake up time (resolved by installing blackout blinds), birds or other city noises waking me up (resolved by getting custom ear plugs), my room being too hot (resolved by using an Eight Sleep mattress topper and an A/C unit), or my usage of digital devices late into the night (resolved by blocking my laptop and phone after a certain hour using Cold Turkey and AppBlock).

Could this work for you?

I think the the main “innovation” that’s made this work for me is setting morning intentions every night, and doing so on a simple piece of paper. In this blog post, I haven’t made any suggestions about what you should do during your routine — there are plenty of other blog posts written telling you what you “should” and “shouldn’t” do in the morning, and I’m not interested in adding to that chorus. What I hope this blog post does is to empower you to design your own morning routine, to make improvements to it on an ongoing basis as you see what does and doesn’t work, and to feel in control of how you spend the first few hours of each day for the rest of your life.

If you try this out, I don’t know if it’ll work for you. This method might be very particular to my challenges with mornings, and the way I think and operate. However, if you do give it a go, I would love to learn what worked and what didn’t work for you! Please consider reaching out, or perhaps leave a comment on this blog. And finally, a huge best of luck! :)

Megaprojects: temporary endeavours (i.e. projects) characterised by: large investment commitment, vast complexity (especially in organisational terms), and long-lasting impact on the economy, the environment, and society.

Megaprojects built the world. We had the first farming communities 10,000 years ago. 5,000 years later, the Great Pyramid of Giza was built. Today, we’re sending humans to mars and spacecraft out of the solar system, and we’re building artificial intelligence that might soon surpass all human capabilities.

How these projects perform has a huge impact on the world, and in some cases, on the future of our species. It’s imperative we improve our ability to strategically identify the need for megaprojects, design them so they generate large societal benefits and minimal harms, and deliver them on time and within budget.

The good

Here are two examples of how amazing humans can be, when we work together effectively towards a positive mission.

The Eiffel Tower

The Paris Olympics are underway, so I’ll start with the Eiffel Tower. It was built in two years and two months, from 1887 to 1889, by <500 people for ~£30M in today’s money. Upon completion, it became the world’s tallest building.

It’s one of the most famous buildings in the world, and is visited by 7M people each year. There are also replicas in Las Vegas and China!

TSMC Fabs

Semiconductor manufacturing is probably the most complex manufacturing process in the world. I wrote a post about it here. It takes place in a semiconductor chip fabrication plant, known as a Fab. TSMC built the most advanced fab in the world (Fab 18) for ~$20B. The chips cost $500M just to design, then TSMC mass-manufactures them with atomic-level precision. They’re essential for every major industry in the world.

The bad

Yes, humanity often does amazing things. However, we’re not immune to blunders, and we have a lot to learn before we can solve all the world’s biggest problems, and create a flourishing future for everyone. Here are some fiascos we can learn from.

Rio de Janeiro 2016 Olympics

The olympics have a terrible track record of going over budget and providing fewer benefits to the host country than anticipated. The Rio de Janeiro 2016 Olympics is one of the worst-performers. Their initial budget was $3B, but by the completion of the games, their costs had skyrocketed by 4x to $13B. While the games led to an increase in tourism during 2016, most of the infrastructure that was built for the games has fallen into disrepair. The economic benefits are dwarfed by the immense costs incurred by the local population.

Berlin Brandenburg Airport

In January 2024, I visited a friend in Berlin, and landed in Berlin Brandenburg Airport. I was dumbfounded by how inefficient the experience was. After landing, hundreds of passengers were crammed into a small space, waiting to have our passports processed. It took ages to get through passport control, and once I was out of the airport, I had no idea how to go from the airport to Berlin. The signage was terrible, and I had to ask multiple strangers and members of staff to figure out what was going on. Google maps was oblivious as to how I was meant to navigate the many layers of the airport.

After sharing my experience with a friend, I learnt that this airport was a national scandal. It was badly designed, opened 9 years late, and had gone over budget by 3x, costing over €7B. It’s now a popular case study of megaproject failures.

Sydney Opera House

Imagine you’re a politician in Sydney, Australia, in the 1950s. Your health is on the decline, you might lose the next election, and you want to establish your legacy. You decide to push for a grand new building that will put Sydney on the global stage, and there’s no time to lose. You succeed in convincing the public and other politicians to invest a huge amount of money into a new opera house, and the design contest has been won by an unknown and inexperienced Danish architect. You can’t wait for the designs to be finished, so you encourage construction to begin. This is how the Sydney Opera House started its life.

Local politicians were incentivised to underestimate the costs and construction timeline, while overstating the expected benefits. They wanted to start the construction phase as quickly as possible, to make it harder for future politicians to stop the project, and they were worried the public would turn against them.

The idea for the shells was revolutionary, and as a result, nobody knew how to build them. They had to make major engineering breakthroughs in order to turn the initial design into reality. Early construction work had to be demolished, as the foundations weren’t strong enough for the final designs of the shells.

There was an election, and due to tensions between the architect and the newly elected official, the lead architect left the project.

The Sydney Opera House was completed 10 years late and went 1,350% over budget. The lead architect, Jørn Utzon, left in disgrace, and never saw the building he envisioned almost 20 years earlier.

What are some reasons why megaprojects fail?

There are many reasons why megaprojects fail, but here I’ll highlight some of the major themes from the book “How Big Things Get Done”.

Impatience leads to mistakes and pivots

We often hear people commend the idea of being “fast” or operating “at pace”. However, if you start pouring concrete foundations for an unfinished building, or you start digging tunnels before you’ve finished the design (looking at you Big Dig), you’re asking for trouble. This often happens due to impatience, external pressures to see “results”, or to generate inertia so no one can stop the project.

Once you’ve started the expensive phase of a project (e.g. construction in a building project, animating and marketing a film in the entertainment business), mistakes or pivots become very expensive. You might discover that your bridge shakes violently in the wind, and you have to spend millions making it safe. You might discover that your movie’s plot is flawed, leading you to scrap half the scenes that have already been filmed. You might discover that nobody wants to use the app you’ve spent 6 months developing in your bedroom.

These mistakes will send you “back to the drawing board”. However, you’ve now contracted thousands of construction workers, or you’ve got a film crew and fancy actors ready to shoot the next scene. You’ll feel immense pressure to find a “quick fix” so these people can get going again. Your quick fix leads to more problems. Years later, the project is finally finished, but nobody is proud of the final result. Your project is over budget, late, and doesn’t meet its promised benefits.

Start with why

Instead of rushing to pour concrete or build a complex app, start by thinking really hard about this basic question: Why?

You should have a very clear understanding of why you’re here at all. What is it you’re trying to achieve? What problem are you trying to solve? What do you want to be different in the world? Only after you have invested the cognitive effort to gain clarity on this should you start ideating on how to achieve the desired outcome.

Learn when it’s cheap

Your first idea is probably bad. You should generate lots of ideas for how to achieve the desired goal, prioritise the best ideas, build cheap versions of them, get lots of feedback, and improve and iterate many times on the cheap versions. Once you’ve tested your idea and you’re confident you’ve got the makings of something great, move to the next stage.

Before you start a time-consuming and expensive phase, you should have thoroughly tested and improved a mockup or simulation of the final product. You should know exactly what needs to be done, and you should feel confident you won’t need to make any major pivots.

For example, if you happen to be building a musical instrument that’s played by marbles, create cardboard cutouts of the instrument before starting to build it. Use the cardboard cutouts to visualise what it might look like on stage, and what it would be like to play. Make CAD drawings of each component, and design a working virtual version of the machine, to identify design flaws before you start welding metal together. You want to avoid spending 4 years building a machine that doesn’t work! (I encourage you to watch all 4 of the videos on the marble machine! What a journey).

If you’re producing an animated film, write a high-level storyline (get feedback, iterate, improve, etc.), then write a script, then create a storyboard, and only once you’ve got an excellent storyboard do you consider starting the expensive and time-consuming animation stage.

Make accurate predictions

You press the ON button on your TV, and a news anchor is interviewing a politician. The politician says “this project will be amazing! It will be completed in 3 years, and will only cost £50M.” Do you believe them?

When proposing a big project, those who stand to gain from it are incentivised to overstate the project’s benefits, and understate the costs and timelines. They want to convince people to get on board, secure funding, and prevent opposition. Their predictions are going to be way off.

To make an accurate prediction of the project’s expected costs, benefits and timelines, try doing reference-class forecasting. This is where you try to understand what category (reference-class) your project is in, e.g. “building a tall, thin skyscraper in New York” or “launching a vegan fine-dining restaurant in London”, and gather data of lots of examples of other projects in that category. You want to graph out the costs of historical projects in that category, how long they took to complete, and the resulting benefits. If everything follows a normal distribution, and you think your project is basically the same as the other projects, then you can assume that your expected costs, timeline and benefits are the average of the examples you’ve found!

Your project is less unique than you think it is. There are always projects you can learn from, that help you ground your predictions. Don’t make up a deadline for your team out of thin air, and then run around telling potential investors you’re going to solve world hunger in the next 2 years with your new iPhone app.

Once you’re going, get moving!

The Eiffel Tower was built in 2 years, and the Empire State Building was built in 1.5 years. They’re both icons. The Berlin Brandenburg Airport took 14 years, and is a national embarrassment.

Once you start the expensive phase of your project, you want to move as quickly as possible.

You want to avoid running payroll for 1,000s of staff for the next few decades! Being slow is expensive, as a result of ongoing staff costs, paying money to rent equipment or land, and increases the delay between starting your investment and making a return.

Being slow can also be very disruptive for the people affected by your project! The Big Dig caused immense disruption to the residents of Boston during the 15 years of construction, generating a huge amount of noise and air pollution, disrupting traffic, and decreasing revenues for local businesses.

You’re also vulnerable to unexpected events during this phase of the project. Extreme weather events, a global pandemic, financial disasters, political turmoil, etc. could stop your project in its tracks, and once a huge project has stopped, it’s difficult and expensive to get it going again. You might lose key employees, your financial or political backing might fizzle away, or the public might turn against you. Reduce your project’s window of vulnerbility by speeding through the delivery/build phase!

How do you move fast?

First, you should know where you’re going! Don’t start before you know what the destination is, and everyone on your team is aligned with that direction. You should also have tested many cheaper versions/mockups/simulations of your designs before proceeding with the expensive “build” phase. Finally, consider building lots of small things instead of one big thing.

Here’s a list of “one big thing” projects: the Olympic Games, nuclear power plants, high-speed rail.

Here’s a list of “many small things” projects: consumer electronics, solar panels, wind turbines, factory-made houses, the Eiffel Tower’s metal pieces.

Before continuing to read, try to come up with a few examples of “one big thing” and “many small projects” yourself.

When you build lots of small, fast and reproducible things, you provide yourself with many opportunities to learn. You can improve the design, manufacturing and delivery of the product with every unit, and your team will become experts in their domains. You want your team to be a well-oiled machine!

You can also isolate the consequences of an error to a single unit. If one unit breaks, it doesn’t destroy the entire batch. And you can start to generate revenue much sooner, without having to complete the rollout of all the units. Each unit can provide value in isolation.

Also, don’t chase shiny objects: use boring technology. If you want to build something huge that’s on time and on budget, use well-refined technologies and methods so you can minimise risky assumptions and find experienced staff. I recommend you read Project Hail Mary so you can enjoy Eva Stratt commanding everyone to save the world using only boring technology!

When you’re building a big, slow, bespoke thing, the learnings you gain on how to do that thing don’t transfer well to your next project. After 10 years of construction, your team might know how to build this type of nuclear power plant in this specific way, but once you move to the next power plant, the science and engineering has improved so the designs have all changed, and your team needs to climb up the learning ladder again, with all the costly mistakes which that entails. You’re afflicted with eternal beginner syndrome.

Concluding

Making the world better for humanity will require many enormous projects. But making big things happen is really hard, megaprojects have a terrible track record, and it’s easy to get swept up in the excitement of a new idea. To increase the likelihood of success, consider following these guidelines:

• Think hard

  • Start with asking yourself why, and work backwards from your end goal.

  • Create mockups / simulations / prototypes of your initial ideas, test them, get lots of feedback, and iterate. Make mistakes and learn when it’s cheap.

  • Do reference-class forecasting to get a realistic estimate for the expected costs, timeline and benefits.

• Act fast

  • Once you’re making big moves, move quick! Big moves are expensive and disruptive.

  • Minimise the duration of time where your project is vulnerable to unexpected events.

  • Scale in a smart way by building lots of small reproducible things, instead of one big bespoke thing.

Before we get to that, let’s first evaluate the default options: I could drive, or get public transport.

Road trip

The drive from London to Anglesey involves staring at the motorway ahead of you for 6 hours while trying your best not to fall asleep. There are a few service stations along the way, where you can get a nice soya latte at Costa and a vegan sausage roll at Greggs. That doesn’t really make up for the monotonous nature of the drive. It’ll set you back £50 in diesel, and £100s in lost life satisfaction.

Public transport

Trains are usually much faster than driving, and you can do interesting things on them like reading your emails. You might even reply to some of them.

The first 30 minutes of my journey is on the London Underground to get to a National Rail Station. I’m now at Euston, and ready to jump on a series of different trains to get to Bangor in North Wales, which will take 3.5 hours on a good day. Once I arrive at Bangor, I’ll wait one hour for the bus, and then sit on the bus for 50 minutes until I arrive at my village. This took 6 hours and cost anything from £50 to £200, depending on the whims of the train pricing algorithms.

Of course, this assumes that the trains aren’t on strike.

Flying car

Did someone mention flying car? Let’s take a trip in a GyroMotion!

I wake up in the morning and walk to a local coffee shop. While sitting at the cafe and taking a sip of a delicious oat milk flat white, I decide to visit my parents for the day. I walk home, jump into the cockpit of my GyroMotion that’s parked outside, and drive for 30 minutes to the nearest airfield. I drive in, say hello to air traffic control, and then take off. The runway is 300m long, but it only takes 70m for me to be in the air.

For the next 2 hours and 20 minutes, I fly over the British countryside. I’m entranced by the beautiful Waddesdon Manor in Buckinghamshire, I fly under Iron Bridge in Shropshire, and I wave at hikers on their way up Yr Wyddfa (Snowdon) in Eryri (Snowdonia). I land in a field outside my house, grateful that the sheep are elsewhere.

The journey from door to door took 3 hours and cost £50 in petrol. After saying hello to my family, I drive the GyroMotion over country roads to the best chippie on Anglesey. I eat the chips while watching the sun set over the ocean.

The challenge

Right now, the journey could actually take a few decades. I’d first need to stop by the UK Parliament and the Civil Aviation Authority (CAA) to do some lobbying. Small aircraft have barely changed over the past half century due to arduous regulations preventing engineers from innovating, and the GyroMotion Calidus isn’t legal to fly in the UK (along with most gyroplanes).

For example, the regulation to determine the “initial airworthiness” of UK aircraft is 758 pages long, and includes sentences like this beauty:

§§ 25.901 and 25.1309 at Amendment 25-40 and 25-46 was adopted in the explicit fuel tank ignition prevention failure analysis requirements of § 25.981(a)(3), the incremental requirement for demonstrating compliance with the ignition prevention requirements of Amendment 25-102 is to develop and implement the fuel tank system airworthiness limitations instead of developing Certification Maintenance Requirements in accordance with § 25.901(b)(2) at Amendments 25-40 through 25-46 and AC 25-19A.

Yes, that was one long incomprehensible sentence. It was the first sentence I read when I scrolled to a random page. The document has 758 pages of those sentences.

As of 2024, only 2 aircraft companies have received approval for any of their gyroplanes to fly in the UK: RotorCraft UK (the British trading name for the German company AutoGyro) and Magni Gyro (Italian). Take a guess when the first gyroplane was invented. Once you’ve made your guess, click here for the correct answer. We’ve made astonishingly little progress since then relative to what we could have achieved if aviation innovators were able to operate with fewer nonsensical impediments.

Where do we go from here

I’ve become convinced that every organisation in the world would benefit from applying this algorithm to large parts of their work. The basic steps, to be done in order, are:

• Question every requirement, and make them less dumb.

• Delete every part or process you can.

• Simplify and optimise. Simple is beautiful.

• Accelerate. Whatever the process or system is, make it faster.

• Automate.

If the CAA applied this to their work, I expect they would end up with far more effective regulations. Vast swathes of the existing text would be deleted, and the documents would be far shorter and simpler. The rules would be focused only on what’s most important for ensuring public safety, while enabling entrepreneurs to develop new aircraft types and improve on existing technologies.

I could then fulfil my Sunday morning dream.

Around October 2023 my symptoms became much worse, and every week or two I’d feel so ill I wouldn’t be able to work for an entire day. I became determined to figure out what was going on, so I started to do a lot more of my own research, having received very little help from any of the doctors I’d seen thus far. I came across the term “IBS,” which stands for irritable bowel syndrome, and the symptoms seemed to line up pretty closely with mine. I immediately went on a “low FODMAP” diet, and saw some fairly good improvements, though it’s also an incredibly restrictive diet and it was not fun.

I decided to reach out to a specialist IBS clinic to seek more advice. I had a series of calls with one of their specialists, the first of which involved discussing all the medical ailments I’d ever experienced, my lifestyle, and my food habits. The intention with this session was to develop a holistic understanding of my gut’s “life experience,” and to start to develop a plan of action for how I could fully recover (and get off the low FODMAP diet).

I went into this experience being pretty skeptical - my instinct is to find something more “data-driven” and receive advice from “real doctors.” However, the experience has been transformative. I’ve experienced basically no symptoms for the past two months, and I’m pretty confident my gut health will continue to improve.

What did I learn?

First and foremost, I learnt that your gut bacteria is pretty important, and it was something I’d entirely neglected. My diet wasn’t varied enough, and I wasn’t consuming enough probiotics (good bacteria) or prebiotics (food for good bacteria). My gut was probably full of bad bacteria, so my food wasn’t getting digested properly.

I also learnt that the way you approach food is also very important. I would often snack late at night, and eat food while working or watching YouTube videos. I was rarely in a relaxed state of mind while eating, and I usually wasn’t even thinking about eating at all.

What did I do to change things?

I now have an alarm that goes off at 5pm to tell me to eat my final meal of the day, and I won’t eat any food afterwards. That gives my gut plenty of time to recover after a hard day of digesting food.

I’m taking probiotic and prebiotic supplements (S. Boulardii, L. Rhamnosus GG, and PHGG) while I start introducing more diverse whole foods into my diet. The main natural sources of probiotics that I’m now eating are yoghurts, kefir, tempeh, and I might explore “kimchi” at some point. Similarly for prebiotics, I’m now eating more oats, legumes, and a wider variety of fruits and vegetables. I’ve started taking Omega 3 supplements as vegan diets are often low in Omega 3, and I’m considering L-Glutamine too.

Finally, I’m drinking a lot more water, and I also add some electrolyte drops to my water bottle every morning.

Where do I go from here?

I’ll continue taking the supplements for a while, and will be investing more time into home cooking, adding varied foods into my diet, and being intentional about eating.

If you think you may have IBS, I would strongly recommend seeking professional support from an IBS specialist, and feel free to contact me if you have any questions :)

Now your toaster can talk to your car.

History

The year is 1989, the cool kids have Sony Walkmans, and mobile phones look like suitcases. The only wireless headphones you can get are bulky AM/FM radio headsets to listen to BBC Radio 4.

The CTO of Ericsson wants to develop a wireless headset that can connect to people’s personal computers. He tasks his engineers with developing a new radio technology for this use case.

Throughout the 1990s, engineers at Ericsson worked on developing this radio technology, and in 1998 teamed up with engineers at Intel, Nokia, Tobisha and IBM to found a non-profit industry consortium called the Bluetooth Special Interest Group (SIG). To ensure interoperability between different devices, Bluetooth was established as an open industry standard, and the SIG was tasked with developing and promoting the standard so that Bluetooth would achieve wide adoption.

In 2000, Ericsson unveiled the first ever Bluetooth-enabled mobile phone: the Ericsson T36, alongside the first ever Bluetooth headset: the HBH-10.

In the year 2023, over 5 billion Bluetooth-enabled devices were sold, and every single electronic device I own that does any kind of wireless data transmission is Bluetooth-enabled (apart from my NFC YubiKey!). Bluetooth has become the global standard for short-range low-power wireless data transmission.

Wireless headphones were the original inspiration for this technology, but Bluetooth is now used in a wide variety of applications. Examples include the Apple AirTags for finding lost items, glucose monitors for people with diabetes, smart lights, and track and trace apps.

So why does Bluetooth still have a reputation of being a bit rubbish? Let’s start with the design requirements.

Design requirements

The original use case was for personal computers to send and receive audio data from nearby battery-powered wireless headsets. These headsets need to be small and lightweight, and must therefore be low-power and have small antenna. The audio quality must be preserved during transmission, with a range of 10-100m as most accessories are expected to be near the user.

Radio technologies benefit significantly from network effects (imagine if you’re trying to connect to your headphones via Redtooth, but your phone only has Yellowtooth), so the founding engineers wanted the technology to be widely adopted. Therefore, it has to be cheap and simple for manufacturers to add it to new devices, and easy and useful for consumers.

Finally, the connection between the transmitter and the receiver(s) must be secure, so that other people cannot intercept, listen in and/or adapt your data transmission.

In summary:

• Low power consumption for battery-powered devices.

• Small antenna for compact devices.

• High-quality audio data transmission.

• Range of 10-100m.

• Cheap and simple to incentivise manufacturers.

• Easy and useful to be desirable to consumers.

• Secure to prevent hacking.

Technical specifications

What technical decisions have the Bluetooth engineers made to satisfy these design requirements?

Bluetooth operates in the 2.4GHz “ISM radio band,” alongside other technologies like WiFi and microwaves. While you need a license from the government to transmit data on most frequencies, the spectrum between 2.4GHz and 2.5GHz has been designated globally as open to anyone to use, and Bluetooth uses 2.40-2.48GHz. Manufacturers therefore do not need to apply for expensive licenses to produce and sell their devices, but they also have to develop ways to reduce interference from other devices operating in the same part of the frequency spectrum. This includes other radio technologies like WiFi, and household appliances like microwaves.

Electromagnetic radiation with a frequency of 2.4GHz has a wavelength of around 12cm. Antennas operate most effectively when they’re (multiples of) a quarter the size of the wavelength, therefore bluetooth antennas can be as small as 3cm. This allows manufacturers to install Bluetooth antennas into very small devices, and the antennas also don’t need to be a straight line so they can adapt to different form factors.

To transmit data from between devices, you have to send out a signal and also be listening for signals. This consumes power. The Bluetooth standard allows devices to transmit with a power of 0.01-100mW, but in most use cases 2.5mW is used. For context, this is 2,000 times less power than a 5W LED spotlight you might use to light up your room. Using this power output and inputting some reasonable numbers for my mobile phone (transmitter) and headphones (receiver) into Bluetooth’s own “range estimator” gives an estimated range of 20m indoors. While this lower power is great for a device’s power consumption, it does significantly limit the transmission range.

The new “Bluetooth Low Energy” (BLE) standard splits up the 2.4-2.5GHz spectrum into 40 “channels” (imagine these as different lanes on the Bluetooth highway), each 2MHz wide. The first channel is 2.402-2.404GHz, the second channel is 2.404-2.406GHz, etc. up until 2.480GHz. One transmitter is able to connect with up to 7 receivers at any moment in time, and they all communicate with each other on a single channel (though they jump between channels all the time to avoid interference, to learn more about that search for Frequency-Hopping Spread Spectrum). Bluetooth is therefore not ideal if you want to connect to a large number of peripheral accessories at the same time using one transmitter, and if there are many transmitters and receivers all using Bluetooth in one location, the channels can become very congested.

BLE has a maximum data rate of 2Mbit/s, so it’d take more than 2 hours to download a 2GB 1080p film. Compare this with WiFi 6 which has a maximum data rate of 9.6Gbit/s, which would take around 2 seconds to download the same film. However, the use case for Bluetooth is for instantaneous small data transmission, e.g. sending quick commands from one device to another, or sending live audio data to a headset. A larger data rate isn’t required for most common use cases.

Finally, who owns Bluetooth? The Bluetooth trademark is owned by the previously mentioned “Bluetooth Special Interest Group,” which is a standards and certification body made up of over 35,000 industry members. You have to become a member of the Bluetooth SIG in order to produce Bluetooth-certified devices, and you must comply with their technical specifications. They also facilitate coordination between different industry stakeholders to improve the underlying technologies and share those benefits with all members. This leads to a somewhat strange situation where the technology is “owned” by a non-profit that doesn’t build any products or do any R&D, but the work done to improve that technology is done by all the independent for-profit companies that are members of the group. An alternative framing is that the technology is collectively owned and developed by all the members, and given the importance of interoperability between devices, there are strong incentives for everyone to benefit from technical advances. However, some companies like Apple have developed proprietary technology on top of Bluetooth (which is why Apple products connect with each other so much better than regular devices).

In summary:

• Bluetooth operates in the 2.40-2.48GHz frequency spectrum, because transmitting data in this part of the spectrum is free and is suitable in terms of antenna size.

• Transmission power is usually 2.5mW, thereby consuming minimal power but providing a small range.

• One device can connect with up to 7 other devices at any moment in time.

• The channels can become congested if there are many other Bluetooth devices in operation nearby, leading to interference.

• The maximum data rate is 2Mbit/s, which is suitable for most common Bluetooth use cases.

Answering the question

So what’s my current hypothesis for why Bluetooth is in most of my electronic devices, but is a bit rubbish?

Firstly, why it’s everywhere: I believe the power of network effects plays a significant role in why the technology standard has become so common - it’s incredibly useful if one device is able to seamlessly communicate and send data to every other device. It’s also been developed with significant industry engagement over the past 2 decades, facilitated by the Bluetooth Special Interest Group. These advances are shared with all members of the SIG via its licensing program and patent pool. One of the SIG’s mandates is also to promote the technology, and I’d guess it’s pretty rare for a technology standard to have an organisation whose job is to promote it.

On why it’s still a bit rubbish: achieving seamless wireless communication between all types of small low-powered devices is an incredibly difficult engineering challenge, and as smartphones have removed the auxiliary port (forcing everyone to use Bluetooth for audio) and WiFi becomes ever more ubiquitous, the traffic on the 2.4GHz ISM Band is only going to increase.

Reflecting on all of this, I’m mainly incredibly grateful to all the engineers who’ve worked on this technology thus far, allowing us to live in a world with pretty good (though not yet perfect) inter-device connectivity.

Perhaps you’ve just listened to an audio version of this blog, using text-to-speech AI and a pair of Bluetooth headphones. If so, I hope your connection was good.

Given the complexity of this process (one could argue it’s the world’s most complex design and manufacturing process!), and that these manufacturing processes are regularly changing and improving each year, this overview is likely mistaken in various subtle ways, so I encourage you to read through the references and do your own research if you’d like a more detailed understanding.

Manufacturing this chip required vast global coordination, intricate supply chains, unbelievable technological innovation, and a decades-long commitment to the continuation of “Moore’s law” by semiconductor chip designers and fabricators.

I’m fascinated by the feats of human ingenuity that led to this, and I hope reading this post leaves you with a taste of this magic. I encourage you to watch the videos that are provided and to look at the diagrams and photos, as they should aid your understanding of what’s going on. Some of the videos are also spectacular (you’ll know it when you see it!).

The following short videos are nice introductions to this topic, that will help you orient towards the rest of this piece.

From sand to wafers

Sand is the building block for the vast majority of computer chips available today. High-purity quartz, a chemical compound of Silicon dioxide, is mined in places like the Spruce Pine Mining District in North Carolina (USA) by companies such as Sibelco (Belgium) and The Quartz Corp (Norway). The refining process includes washing the quartz with chemicals and air (called “flotation”) to remove impurities such as feldspar; magnetic separation to remove ferrous contaminants; acid leaching to dissolve other impurity minerals; and thermal processing (calcination) to remove residual organic compounds. The purity of the quartz for semiconductor manufacturing has to be >99.999%, which costs around $5,000-10,000 per tonne to mine and refine. The crucibles which house the silicon metal during the next phase of production have to be made of the purest form of quartz to avoid any defects and chemical reactions, and quartz of the requisite purity for these crucibles is primarily found in the Spruce Pine Mining District.

High-purity quartz is usually then exported to Japan, to be processed into silicon wafers by companies such as ShinEtsu and SUMCO (both Japanese), ready for chip fabrication. These companies take the quartz and reduce it to silicon metal, which is then processed further into an atomically perfect silicon ingot, via the Czochralski method.

Silicon wafers are produced by slicing these ingots, polishing them to ensure they are (very) flat, before a final inspection and sale to a semiconductor fabrication plant (aka “chip fab”, “fab” or “foundry”).

Building integrated circuits on a wafer

A silicon wafer by itself doesn’t have much value. In order to have commercial applications in the electronics industry, fabs “build” integrated circuits by “printing” billions of transistors and electrical connections on the wafer surface. This is done by carefully depositing thin films of materials onto the wafer, and using UV light to create specific (nano-meter scale) patterns in those materials. The patterns are determined by the chip designers, and those designs must be converted into photomasks that block certain parts of the wafer from being exposed to UV light, in order to create the patterns. This process of using light to produce patterned films of materials on a silicon wafer is called photolithography, and is explained in more detail below.

The company responsible for turning silicon wafers and other materials into Apple M1 chips is the Taiwanese Semiconductor Manufacturing Company (TSMC). TSMC is one of the world’s leading semiconductor manufacturing companies, fabricating a significant fraction of the world’s most advanced chips. The fabrication plant for the Apple M1 chip is TSMC’s leading fabrication plant, Fab 18. It’s located in Taiwan, and is estimated to have cost around $20 billion dollars to build.

Wafer preparation

Before a wafer is ready for the “printing” stage, they must test and prepare the surface for later processing. They inspect the wafer for defects, subject it to a thorough cleaning process, polish the wafer to the appropriate level of smoothness, and confirm the wafer is the correct thickness. Finally, to provide an interface between the wafer substrate and the layers that are printed on during later stages, they grow a small layer of silicon dioxide on the surface by heating the wafer in a furance.

Deposition

To form the desired layers and structures in an integrated circuit, thin films of materials such as metals, insulators and semiconductors are deposited onto the wafer surface. Metals like copper or alumnium are deposited to form interconnects, which act as wiring to electrically connect different components. Semiconductor materials such as silicon, germanium or gallium arsenide are used to create the transistors, which allow the amplification or switching of electrical signals within the chip. Insulating materials such as silicon dioxide are used to insulate different components from each other, to prevent electrical leakage.

Chemical vapor deposition is a common deposition technique, whereby the wafer surface is exposed to an easily vaporised starting material, which then reacts with or decomposes onto the surface. This process is conducted under high vacuum. Patterns are then etched into these materials via the lithography process described below.

Photoresist

Photoresists are materials that change characteristics when exposed to various wavelengths of light. They’re used by fabs to create the 3D patterns on the wafer surface, by coating the photoresist onto the wafer surface and exposing only certain parts of the photoresist to UV light. Once the photoresist has been applied via a spinning process, the wafer is “soft baked” to solidify the photoresist (and evoprate any remaining solvents, from previous photoresist applications). Most fabs apply a “positive” photoresist coating, which deteroriate when exposed to UV light, and can then be removed with a solvent to leave the desired pattern on the wafer surface.

Photomask

Once the chip designers have sent their designs to the fab, those designs are converted into photomask (or photoreticles), hereafter “masks.” Masks are the “blueprint” of an integrated circuit design, as they determine what patterns and circuits are developed on the silicon wafer. They’re usually made of highly transparent quartz glass, with a chrome stencil that prevents certain parts of the wafer from being exposed to UV light.

To print the same pattern across the wafer, the wafer is “stepped” from position to position until the entire wafer has been exposed to UV light through the mask. A single chip design may require a set of 10+ different masks to provide the instructions for the entire printing process, and the production of the blank photomasks is also a significant technical challenge due to the transparency required. Zeiss (Germany) is one of the leading manufacturers of blank photomasks.

Photolithography and EUV

Very low wavelengths of light are required to create the intricate features required for cutting-edge chips. An analogy here is that an artist needs a very fine pencil to draw intricate features in their artwork, and similarly, fabs require very small wavelengths of light to “draw” or print intricate features on the silicon wafer to create functional transistors. As a result, the wavelengths of light used during lithography have been trending downwards: in the 1970s and 1980s, visible light (400-700mm) was used, and by the 2000s, most semiconductor manufacturing used Deep Ultraviolet light (193nm). As of the early 2020s, the frontier is Extreme Ultraviolet (EUV) light, which has a wavelength of only 13.5nm.

To generate EUV light, one cannot use a regular bulb. Reliably creating EUV light is an immense technical challenge, which has been in the works since at least 1984, but it wasn’t until multiple breakthroughs at AMSL (Netherlands) in 2018 that it became commercially viable, through their new EUV lithography systems. TSMC purchase EUV lithography machines from ASML for a staggering price of $150M per machine.

Vaporising tin with lasers

ASML’s EUV machines use CO2 lasers produced by TRUMPF to create ultra-high powered 40kW laser pulses, easily strong enough to cut through steel. Molten droplets of tin of around 25 microns in diameter are then dropped into a vacuum chamber, where they’re initially hit by a low-intensity laser pulse to turn them into pancake-shaped plasma, followed by a “full-power” laser pulse which heats them up to 220,000 degrees Celsius and vaporises them, emitting EUV light. 50,000 droplets of tin are released into the vacuum chamber each second.

World’s best mirrors

EUV light is absorbed by almost all materials, including air, so the EUV light column emitted by the vaporised tin must remain in high vacuum. Lenses cannot be used as they would also absorb the EUV light, so several mirrors are used to guide the EUV light through the mask and to the wafer, where the mask’s pattern is magnified by a factor of four.

Zeiss produces these mirrors (and the overall optics system), and these mirrors are the flattest objects humanity has ever created. If they were the size of Germany, the tallest mountain would be just 1mm high. They’re produced by depositing around 100 (nanometer-thin) layers of Molybdenum and silicon on top of each other. Mirrors composed of multiple thin films of dielectric material are also known as dielectric or Bragg mirrors. Precisely targeting this light onto the silicon wafer is also an immense technical challenge, requiring sensors that have nano-meter scale precision. Zeiss boasts that due to their mirror’s tilt stability, “if one of these EUV mirrors were to redirect a laser beam and aim it at the Moon, it would be able to hit a ping pong ball on the Moon’s surface.”

Hard bake and development

Once the photolithography stage is complete, the wafer is immersed in a developer solution to dissolve the exposed photoresist. The unexposed photoresist remains on the surface at this stage, in a pattern corresponding to the blueprint in the photomask.

The wafer is also “hard-baked” to harden the unexposed photoresist, to be resistant to subsequent processes (etching, deposition and ion implantation), before being removed entirely later.

Pattern transfer

To transfer the pattern into the substrate (or a thin film deposited above the substrate), pattern transfer techniques are used. The most commonly used pattern transfer approach is etching, which involves using gases (dry etching) or chemicals (wet etching) to selectively remove material from a thin layer on the surface of the substrate. The remaining photoresist “resists” the etching process, protecting the underlying material, whereas the unexposed substrate or film is “etched away”, leaving the desired pattern in the underlying layer.

Deposition is another pattern transfer technique. One deposition technique is atomic layer deposition (ALD) aim to deposit lay-ers of individual atoms onto the wafer surface. Meanwhile, selective deposition techniques (such as “Area-selective ALD”) aim to deposit layers of atoms in exact places on the surface, as opposed to covering the entire exposed surface. Deposition at this stage of the manufacturing process is often done to place copper interconnects onto the chip.

Finally, ion implamantation is a way to change the conductive properties of the wafer via the controlled addition of dopants. This is done by accelerating ions such as boron, phosphorus or asenic towards the wafer, which penetrate the surface.

Photoresist removal

Once the desired pattern has been transfered to the substrate or a film above the substrate, the remaining photoresist is removed. This is primarily accomplished using “plasma ashing”. Plasma ashing involves ionizing a gas such as oxygen and exposing it to the wafer. This gas is highly reactive to the photoresist material as it’s an organic material, but leaves the (inorganic) material below the photoresist untouched.

And repeat!

Most advanced chips today are comprised of many layers (up to 100!) stacked on top of each other with nanometer precision. To build these layers, the steps from deposition to photoresist removal are repeated to produce each new layer, which may serve many different functions within the chip.

Die Separation and Packaging

A single fabricated wafer may contain around 200 Apple M1 Pro chips, given the wafer is 300mm in diameter and the M1 Pro chip’s die size is 245mm2. The finished wafer will also include predetermined thin (<100µm) “scribe lines”, representing which parts of the wafer must be cut, to allow the separation of each individual chip. The wafer is bonded to a mounting tape to ensure the chips remain in place during the separation process, and then a diamond tipped saw is used to cut the wafer along the scribe lines. Laser and plasma saws are also used.

Once the chips have been separated from each other, they’re packaged to assure protection from physical damage and corrosion, and to provide a reliable means of interconnection to a circuit board.

Wafer fabrication suppliers

The production of cutting-edge semiconductor manufacturing equipment is dominated by a small number of companies, namely Applied Materials (USA), Lam Research (USA), KLA Corporation (USA), Oxford Instruments (UK), Samsung Electronics (South Korea), and Tokyo Electron (Japan). They build equipment for fabrication steps such as chemical vapor deposition, etching, polishing, ion implantation, and wafer inspection.

The chemicals and materials for semiconductor manufacturing, such as the photoresists, solvents, and etchants, are primarily manufactured by Fujifilm Electronics Materials (Japan), Dow Chemical Company (USA), and JSR Corporation (Japan).

Building the EUV machines required for the lithography of cutting-edge chips requires 1,000s of different suppliers and organisations, which are orchestrated by ASML (Netherlands), which also assembles the machines. No other company in the world can produce these EUV machines, and it’s taken ASML decades of R&D and partnership-building, and billions of dollars in investment, to get to where they are today. As previously mentioned, the optical systems for ASML’s EUV machines are developed in collaboration with ZEISS (Germany), while the lasers are produced by TRUMPF (Germany).

The complexity of this supply chain is absurd, and hence cannot be adequately summarised or described in this post. You can learn more from this CSET report, in the book “Chip War”, and in this report by the Seminconductor Industry Association.

Concluding thoughts

I initially set out to spend one day researching and writing about semiconductor manufacturing, and to post a piece on my blog that same day. That was in January, and it’s now April. Eventually, I spent over 100 hours researching semiconductor manufacturing, and I’m still finding entirely new processes, systems and equipment during my continued research, so I’ve probably made some mistakes or missed important details in this piece.

Nevertheless, I hope this piece has helped you develop an initial understanding for chip manufacturing, and given you a flavour for the persistent human ingenuity and global collaboration required to create the digital world we now live in.

References

If you’d like to investigate this further, here are some of the resources I found useful! Contact me if you’d like access to my research notes and other resources.

Articles, reports, and blog posts

• Semiconductor Technology from A to Z - Philipp Laube

• Semiconductor Lithography (Photolithography) - The Basic Process - Chris Mack

• Knowledge Centre - Semiconductor Engineering

• Lithography principles - ASML

• CHIPS for America - NIST

• 3nm Technology - TSMC

• ASML-Zeiss, a successful partnership enabling Moore’s law - Zeiss & ASML

• Supply Chain Explorer - Emerging Technology Observatory

• Making Chips At 3nm And Beyond - Semiconductor Engineering

Podcasts

• ASML, the Obscure Powerhouse at the Cutting Edge of Chip Technology - Odd Lots

• TSMC - Acquired

Videos

• How does a chip go from design to mass production? - ASML

• TSMC Analysis (playlist) - Asianometry

Back in the good old days of the 1990s, when coronavirus was just a funny word, I was born into a Welsh-speaking family in North Wales. Shortly afterwards, there was a historic referendum in Wales, asking the people of Wales whether they believed Wales should have an Assembly. By a meager 6,721 votes (50.3 to 49.7%), Wales declared it wanted an Assembly, and Cynulliad Cenedlaethol Cymru (National Assembly for Wales) joined me in its infancy. A desire for self-government and independence was on the rise in my home country, and it was in this context I grew up.

When I was around 9 year’s old, I regularly went around all the computers in my primary school and changed their search browsers to “gwgl.com” - the Welsh google page. Those agrieved by its unfamiliarity would frequently change it back to Google, but I was relentless in my quest. At this young an age, I felt compelled to spread Welsh everywhere I went and impose it upon others. However, some of my more nationalistic teachers still didn’t appreciate my efforts. Towards the end of primary school, I was accused by a teacher of betraying Wales and my ancestors for having misbehaved during a class on the Celts. I can still feel 10 year old Dewi’s shame. This is clearly ridiculous, but affected me deeply at the time.

During my time at high school, I became even more patriotic, and felt strongly that everyone in Wales should be speaking Welsh. I couldn’t believe that so many of my Welsh-speaking friends were speaking English with each other, and that Welsh-speaking parents spoke English with their children. This was unfathomable to me. My mind could not process that people would have different passions and motivations in life, and didn’t actually care which language they used to convey their feelings to each other. I also felt (and still do feel) that they were missing out on so many lovely features of our community, that was done through the medium of Welsh and with the Welsh language as the glue that kept us together.

The neglect of my peers towards Welsh culture strengthened my feelings of anger against “the English,” who in my mind clearly led to this despicable situation by hypnotising all the Welsh people with their Welsh Not, Brad y Llyfrau Gleision, the death of Llywelyn ein Llyw Olaf, etc. etc. I also felt incredibly depressed about the future prospects of the culture I loved, as it seemed all but doomed to me. I was utterly failing to inspire some of my friends to speak Welsh with each other in school, and more and more English was heard on the playground.

Towards the end of high school, I started learning about politics, and opened my eyes to the wider world outside of little old Wales. I remember sitting in the computer room and having to research the political parties for some project or other, and being shocked to realise that I knew absolutely nothing about UK politics, who was in charge, what they believe, what their track record was, and why everyone around me seemed to hate the current government. During this time, a friend of mine mentioned to me that he thought my thinking on “the English” was pretty problematic and verging on racism, and I also met what I can only describe as a Welsh extremist who also called me a “traitor” for considering going to university in England, and all of a sudden everything I had previously felt so strongly about came into question in my mind. For the first time ever, I asked myself: why do I believe this?

The romanticised past I yearned for, the purity of a Wales without English, the eradication of those who had oppressed us, suddenly made a lot less sense to me. Without a doubt, the UK Government (and British companies) have done some absolutely terrible things all over the world, from the slave trade to the Bengal famine and beyond, but that wasn’t what was driving the nationalist sentiment around me and within me during my teenage years. We learned at school how bad the English had treated us, the Welsh, and I felt we needed to do something about it. This was the problem I felt compelled to solve: avenging my ancestors and gaining independence for Wales. It wasn’t that we were being taught that the current system was systematically oppressing Welsh people, it was the long history that carried with it so much emotion that was driving our anger towards today’s government. (A small disclaimer here that this does not reflect everyone’s experience of education in Wales. I’m sure many people didn’t have overtly nationalistic teachers or just ignored these types of lessons.)

Then I learnt about climate change and biodiversity loss. What a shock that was. What did it matter if Welsh was going extinct, if the human race would get their first, potentially the only source of intelligence life in the entire universe, along with the rest of the animals on our planet? This was the first time in my life I really appreciated perspective, scale, and importance. Depending on what your perspective is (what you love, where you’re from, what your parents tell you, etc.), you will often have completely different beliefs about what is important, regardless of the actual scale of that problem and the amount of suffering it causes. I realised that the decline of my language was not actually causing that much suffering in the scheme of things, but climate change and biodiversity could ruin us and was already leading to huge suffering in poorer countries. I did the only thing I could think to do: I put down my revision notes in the middle of my A Level exams, and spent four days in my garden building a pond. Maybe not quite the most cost-effective intervention, but hey, we’ve all got to start somewhere.

A few months later, I rocked up at Durham University for the first time, having just been on a two week mad-one visiting my friends in all the main universities in Wales and drinking a ridiculous amount of alcohol. This continued for some time in Durham, and I’m sure many of you who were with me in first year at Chad’s can attest to the madness. At Chad’s, we were blessed to be fully catered during first year, and we’d eat together for all our meals. Here, I met a very strange species for the first time in my life: omnivores-turned-vegetarian. I relentlessly interrogated many of my vegetarian peers, trying to understand why they would sacrifice the opportunity to eat those lovely chicken breasts. Some thought it was healthier. Some didn’t want to hurt animals (”but they’re just farm animals!” I thought, assuming farmed animals weren’t worthy of moral consideration). Some said it was for the environment. It was this final point that niggled my conscience.

I spent a few months after turning 19 researching this point further, trying to understand the link between animal agriculture and environmental destruction, both through climate change and biodiversity loss. It turns out my tofu-eating friends were correct: eating meat does indeed cause much more planet destruction than eating the equivalent plant-based meal. I’d never eaten a vegetarian meal in my life, and was quite a fussy eater, so I was in a bit of a quandary. My values, as a self-described environmentalist, were in conflict with my way of life. I bit the bullet, and asked a friend of mine to join me in eating no meat for a week. That week turned into “until exams are over, then we’ll go for a lovely raw steak.” That steak was never consumed.

That summer of 2016, after my first year at university, I went on a charity expedition to Borneo “to save the orangutans and the turtles.” It was here my thinking really developed. I read a book by the moneyless man, Mark Boyle, and came to the conclusion that money was the source of all evil. Down with the capitalist system! (Disclaimer: I no longer believe this, there is very little nuance in that perspective.) I also witnessed immense suffering at the wildlife sanctuary we were volunteering at, including a crocodile whose face had been ripped apart but was forced to continue to just about survive in a small cage as a zoo spectacle, and I started to question myself again: why am I vegetarian just for the environment, if eating meat also causes billions of animals to suffer every year? And what about eggs and milk, don’t they also cause suffering? I pledged to go vegan, and started to advocate for all those around me to stop eating meat too. Suddenly, seeing people eat meat became morally outrageous to me, in the same way that I used to sit in disbelief as I listened to my Welsh-speaking friends speak English with each other. Why were my friends and family still eating meat, when it was causing so much destruction to our planet and to our descendents, and so much suffering to animals today? This is something I still ponder regularly.

Upon returning to Durham in October 2016, I joined the vegetarian and vegan society, stopped eating all animal products, and started walking around town in socks and sandals. I clearly had to live up to the stereotype. I had co-founded a charity toastie bar with a friend of mine in college before I went vegetarian, the friend who pledged to go vegetarian with me for the months before exams, and we were still serving meat at the toastie bar upon our return to university. This niggled at me for quite some months, as I was not only going out and buying meat for our customers, I was also forcing many of the other chefs who were vegetarian to handle that meat. The charities we were supporting were also animal welfare charities, and this seemed ludicrous to me that we were doing all this work to raise funds for animal welfare while giving students ham and cheese toasties.

One day in April 2017, we decided to get rid of all meat from the toastie bar, and turned it into a vegetarian toastie bar. We consulted no one, and basically forced it on the college. It was not our brightest move, and it led to months of drama in college. During this drama, my friend and I repeatedly stood up in front of the whole college and made the case for the toastie bar being vegetarian. We had a lot of pushback, and we made some crazy speeches, involving dying horses in Patagonia, the end of life as we know it, selling our organs on the black market, and offsetting murder by donating to Oxfam. Remarkably, we emerged victorious, with 95% of the college supporting us. An awesome feeling.

So by this point, I’d gone from being a Welsh nationalist to an environmentalist to an animal rights person. I still cared about my Welsh culture and the environment, but the billions of animals that are suffering in factory farms was my greatest concern. It was around this time a friend of mine suggested I watch a TED talk on a funny-sounding movement: “Effective Altruism.” I had no idea what altruism meant, and the “effective” part sounded pretty pretentious, but I watched it anyway. “Wholly molly!” was my response.

Effective Altruism (EA) is a social movement that’s trying to determine what the best things we can do to improve the world are, using evidence and reason. It was exactly the community of people I was looking for, and I became obsessed. EA introduced many many new things to me, including rationality, frameworks for determining how important various “cause areas” are, huge amounts of resources for better understanding the world we live in and all its uncertainties, and a path forward for what I can plausibly do to contribute to actually having an impact in improving the world. EA is a very young and changing movement, always adapting to new information and better arguments, so it’s very dynamic and can’t really be pinned down to any stationary beliefs, but I’ll provide my current reading of it below.

People in the EA community want to maximally help make the world a better, happier place, with less suffering. This includes all humans alive today as well as all future humans, and any animals that can plausibly suffer. The current areas of greatest focus, based on the currently used framework of “scale” (how big of a problem is it), “neglectedness” (how little attention does it receive globally), and “solvability” (can we actually do anything to improve the situation), are:

• Existential risks. These are things that could lead to the end of the human race such as extreme climate change, super-deadly pandemics, and uncontrolled emerging technologies like AGI. Preventing these risks from killing us all helps ensure that trillions of future happy humans can exist, and I doubt all of humanity dying would be an enjoyable affair either. Somehow, this field of research has almost entirely been neglected until recently.

• Global health and poverty. Around 8% of the world still live in extreme poverty, without acceess to basic healthcare and nutrition, and billions of people’s lives could be massively improved with more access to basic resources that we take entirely for granted in the UK. “Solving” this might be hard, but it doesn’t mean we shouldn’t try, as well as do a lot more research into what is most likely to actually work.

• Animal welfare. Over 70 billion animals are slaughtered every year for our animal flesh habit, and these animals almost certainly live terrible and suffering-filled lives. It is more efficient environmentally in terms of land use, water use, and co2 production to not eat meat, and yet we still inflict this unnecessary suffering upon these animals. Technologies like plant-based and cultivated meat seem most promising in ending this problem at the moment.

• Promoting research and cooperation. There are enormous uncertainties surrounding the above, and it’s likely there are even greater problems facing the world that we haven’t even thought about or haven’t emerged yet. Having better philosophical foundations for tackling these questions, more research into our priorities and what risks we face, and better mechanisms for making decisions and working together are likely to make the world’s inhabitants much better off in the long-term.

• This is not exhaustive! Learn more at the 80,000 hours website.

It is in this community I now find myself, presently as president of EA Durham, and I continue to be inspired by and learn from its many members worldwide. The world is far more complex than I’d previously appreciated, but there are also many things we know with high certainty would make the world a significantly better place, so finding and implementing the most effective ways we can bring about those changes is an awesome and inspiring challenge for all of us. It’s likely my thinking will continue to evolve as I learn and think more, and EA as a movement will almost certainly change with time, and I’m excited for that journey. While I’ve been criticized in the past for “changing my mind on big issues,” I still think that updating your beliefs after learning new information, backed by rigorous scientific evidence from reliable sources that you’ve deliberated over thoroughly, is something we should embrace, while pushing back against ideological silos that narrow our understanding of people and the world.

One outcome of my involvment in EA has been that I’ve taken the “Giving What We Can” pledge, to donate at least (and hopefully much more than) 10% of my lifetime income to effective charities or organisations working to improve the world and its inhabitants. The top 10% of my income is very unlikely to make my life much better, but could for example add enormous value to someone’s life in a poorer country. If this isn’t something you think you can do, have no fear, it’s by no means a prerequisite to getting involved with Effective Altruism, although it is a very easy thing we can all do to increase the good we do in the world.

I’m now at the juncture where I’ve almost finished education, and have to decide how to start the rest of my life, and what path I want to follow or create for myself. I’ve always been “mission-oriented,” wanting to solve certain problems, and I’m so grateful for all the work people in the EA community have done in breaking down the question of what the biggest problems in the world are and where we can have the biggest impact. Now all I have to do is decide which of these many problems I’m most passionate about, and how I can contribute most effectively using the skills I have and could develop. If only this was an easy question. I’ll keep you updated.