I was about eight when we acquired our first five transistors. They were in a portable AM radio that didn't need to warm up and ran on a 9 volt battery. It was something of a revelation to me. Up until that time the only portable radio we had wasn't terribly portable. An ancient box with four vacuum tubes, it required two different batteries - a 1-1/2 volt A battery and a 67-1/2 volt B battery. These were expensive and hard to find, so running the radio meant using its electric cord. The closest thing we had to portable was the five tube Philco radio in the Ford Falcon.
It was possible to count the key components of home technologies back then - the number of electric motors, vacuum tubes, transistors - as everything was discrete and physically large. As I got a bit older I learned what each was for and could wire up basic circuits eventually making some of my own design. Learning how they worked was the gateway drug to amateur radio just as playing with small microprocessors, sensors and code is a gateway drug to possible futures now.
Unknown to me at the time people were beginning to work on changing the world. The transistor had been invented at Bell Laboratories in Murray Hill, NJ with the intention of improving long distance telephony. To maintain its monopoly status AT&T made a deal with the government that made inventions from the Labs generally available. Bell Labs could work in areas that impacted telecommunications, but could also follow paths that branched away as it had become an invention engine for the country. New companies sprouted - notably Fairchild in Silicon Valley - and new branches of invention and innovation emerged. By the mid 60s a silicon revolution was well underway.
About 50 years ago Gordon Moore published his famous paper outlining what has become known as Moore's Law - basically a projection that a dramatic improvement in the performance of silicon based electronics would extend for a long time. It didn't come from a vacuum. Moore was working on putting multiple transistors on a wafer of silicon and wiring them up - early ancestors of what we would call integrated circuits. It was clear the elements could be made using a lithography process and that shrinking was possible. The question was how far could you go.
Moore was at Fairchild but had a strong connection Carver Mead at Caltech. Mead was interested in the physics of silicon transistors and was able to work out that it should be 'easy' to shrink individual transistors to about 150 nanometers without running into fundamental problems. This was something of a staggering result for Moore - Mead's numbers were more than 100 times smaller than anyone thought possible.
Mead was interested in fundamental limits rather than those that were limiting manufacturing. Moore's conjecture was that the lure of greatly improved performance and lower prices was enough to fund whatever it would take to solve the technical problems of manufacture along the path. At its core Moore's Law is a fundamental belief that sufficiently motivated people can be very clever and accomplish amazing things.
Fifty years, trillions of dollars, and an enormous amount of human effort and we're below Carver Mead's conservative limit by about an order of magnitude. There are certainly limits and perhaps it makes more sense to think about the amount of power required for computation, but fifty years is a good point to sit back and take stock of the progress made to date.
I have a iPhone 6 Plus - the gold 64 GB version. A quick back of the envelope tells me it has slightly less than 290 billion transistors - quite a jump from five in that old transistor radio.1
My iPhone has roughly the same number of transistors as our galaxy has stars.2
There are a lot of transistors in the world. An estimate by Dan Hutchenson VLSI Research suggests 2.5 * 1020 transistors were fabricated in 2014.3 That's about eight trillion per second.
Moore's Law slowed down a bit about ten years ago. There is still progress, but other metrics like average power consumption per unit of computation are becoming centrally important with mobile.4 We may complain about the lack of progress in batteries - most technology does not follow the same steep slope of Moore's Law - but there have been improvements...
I think of A and B batteries and smile...
1 The vast majority of the transistors are in the memory chips and A8 processor. Other chips may have up to a few million transistors each, but ignoring them turns out to be a rounding error. The flash memory is MLC NAND - each transistor can store two bits of information or four transistors per byte. My iPhone has 64 gigabytes, we we're talking 274.88 billion transistors (remember a kilobyte is 1024 bytes) for the flash. It has 1 GB of SDRAM so another 8.59 billion transistors as SDRAM uses one transistor per bit. Finally the A8 processor is advertised at over 2 billion transistors. So something north of 285.8 billion considering other components - probably just under 287 billion transistors.
Also note that something like a smartphone has about 99% of its transistors devoted to memory.
2 Current estimates of the number of stars in the Milky Way galaxy are in the 200 to 400 billion range.
3 I don't have access to the input numbers. You can do some crude checks. The VLSI Research number gives about 36 billion transistors per person. This works out to about 100 billion transistors for the 2 billion or so connected people. With smartphones obsoleting in about two years. An average smartphone (including the cheap ones) probably averages something under 100 billion transistors so we're in the ballpark.
The increase in transistor fabrication has been so fierce that you only have to consider the past 3 or 4 years before earlier numbers are become rounding errors - also going back much further and the devices are no longer in use.
4 It is well beyond the scope of a quick post, but Moore's Law had at least drivers in addition to human ingenuity and will. One was the extreme scalability of photolithography. The tools may seem expensive, but they've been able to generate increasingly vast and inexpensive output even as designs shrunk and became more complex. And there is Dennard scaling. In the mid 70s it was discovered that the current and voltage required to switch transistors shrank with their size. Without this chips would have quickly melted down and defied any shrinking attempts.
Currently something of an explosion of ideas is underway of where we go beyond Moore's Law. The rough principal still applies - devices will become more powerful in some meaningful way - just not exactly what Moore conjectured.
I made a chocolate cake to celebrate the birthday of a vegetarian friend. It is based on an eggless chocolate mousse. The trick is to use avocados. I won't go into the full cake other than noting the first layer is a shortbread with a lot of cocoa powder. The chocolate mousse goes on that and everything is covered with a chocolate glaze and then a semi-sweet chocolate is peeled over it (anything darker than about 50% doesn't peel cleanly).
The chocolate mousse recipe is about right for a 7 or 8 inch round pan. I used a small 5 inch pan and made several. Adjust as you see fit
° 2 ripe avocados
° 100g dark chocolate, melted (I used 70% Lindt)
° 2 tablespoons dark cocoa powder
° 100g white cane sugar
° 1 teaspoon vanilla extract
° 1 pinch salt
° 1 tablespoon coconut oil
° 3-4 tablespoons coconut milk
° Combine all the ingredients in a food processor and pulse for 5 minutes or so until smooth.
° Add coconut milk as needed to thin is down.
° Spoon the mousse over the base and level
° Refrigerate the cake overnight.
° The next day, remove the cake from the pan and transfer it on a platter.