Analog Computing is GENIUS - Here's Why!

Check out the Roborock S8 Pro Ultra Robot Vacuum Today! amzn.to/3v5Yftf

My first experience of digital computing was in 1962 when the U.S. Navy trained me for their first three ships at sea to be equipped with them. While those early computers were much smaller than an IBM mainframe of the time, they were the size of large double door refrigerators, and needed chilled water to cool the processors. Energy costs were high despite the transistors they used, and processing times were much slower than the simple digital devices we use today. The digital computers were used as an air traffic control system, for combat and it fed some of the information about potential targets for our missiles to the fire control technicians on a different set of display consoles. Their computers and consoles were analog. At that time the speed of their analog resolvers and differentiators was superior to any digital simulation our computers could produce. But they were constantly having to fine tune the analog elements to their system to maintain accuracy. As you note, digital can be very precise and repeatable.

I started my career working on analogue computers used in industrial and manufacturing environments. They were mostly used for controlling motors and used proportional, integration and differential calculations at very high speed. They worked well in environments that were not computer friendly and were very robust and reliable. That was 40 years ago and they lasted for years because they just worked. I think the challenge was that to program them you needed to understand electronics and mathematics and that isn’t so necessary with modern digital computers.

Having grown up in the golden age of record players I fall on the side of digital. After a single play on a turntable a record will lose that frequency response curve. It really doesn't matter how expensive the turntable the wear and tear on the record is a function of the design. A physical needle riding grooves wears a bit of those grooves away with each play of the record. The quality of the digital recording is there forever.

you know, in 1974, researchers had developed an analog matrix method that could recognise the alphabetic and numeric characters, individually, with a very high success rate, and i remember being surprised at thesmall amount of analog hardware to perform the job, once learned/programmed...and, even then, the analog methods were very fast and very low power, per "recognition"

Love learning about diverse computing methods. Between analog, digital, and quantum computers, we'll be able to solve problems so much better than ever before!

I had a little trade school on automatic transmissions of the 1980s and before. I always asserted that the valve body in one of those transmissions was essentially an analog computer that operated hydraulically instead of electronically. I still assert that.

Yeah, Digital was never truly Digital you know, it was always just some conventions about analogue voltage levels(TTL/CMOS), and in a usecase like this, storing neuron BIAS values "vertically", we're basically removing those conventions and unleashing the full potential that every "digital Bit" had in the 1st place. LOVED this video... it suited my old school analogue brain really well 🙂

I remember using mechanical computers in the Navy that measured speed of targets for our Fire Control systems. Those computers wear far faster than anything around at the time.

I just realized that analog computers with a digital nanny to adjust for drift or runaway errors is how the fuel control unit in many modern jet engines operates. They use an analog hydromechanical computers to calculate and dispense the fuel as required, and a digital computer to adjust things on the fly to keep everything optimized for the best balance of fuel economy, longevity, power, and emissions. Many small aircraft still use what is essentially a purely analog hydromechanical computer to control the fuel flow to the engine. The fuel injection on these piston engines uses inputs such as throttle position, fuel lever position, airflow, RPM, and air pressure to compute and meter out the fuel necessary for those parameters. Every annual or 100 hour inspection includes checking the fuel mixture at high and low power settings and adjusting as needed to keep the fuel system output correct.

Great content today and always. I am somewhat of a music snob and grew up in the analogue era. I embraced the digital movement in music as it was introduced and find digital very satisfying. My ears can only hear so much frequencies as many of us baby boomers grew up in the rock n roll era . Therefore as close as digital simulation is getting to analogue it is good enough for a large percentage of the population's ears. Thanks again and keep the great content coming

Another pathway to solve the energy efficiency issue would be classical reversible computing. The main tricks used by the designs explored in this video have to do with the fact that you need less bits of informations to describe a physical system when accuracy isn't critical AND you spend much less energy when this less accurate memory is closer to the computing units (in proper neuromorphic computing, those tend to be the same, even). Reversible computing directly goes after the fundamental cause of energy dissipation in digital systems (which, you have to remember, are necessary for faithful information transmission and general error-correction) aside from intrinsic factors such as friction and electrical resistance (that analog computers don't address de facto either): information erasure. It turns out, from Landauer's principle that there is an irreducible energy cost associated with information erasure at the bit level and since the logic gates used in conventional CPUs aren't reversible/injective (simply for cardinality reasons), these costs add up... before you even start taking intrinsic resistance into account. Even a perfectly superconducting, ultrastable conventional CPU would need to dissipate tens of Kilowatts of heat to perform a ZettaFLOPS of compute. In contrast, there are reversible nanomechanical "chips" designs that could theoretically perform at the Zettascale while barely dissipating a single Watt of heat... less energy than a typical smartwatch ! More than analog computing, it's the field of reversible computing that we should keep an eye on.

From what I've read, Mythic's tech isn't really quite analog. One source reported 256 descript gate voltages for each transistor storing a value. This is no doubt closer to analog than digital binary states are, but the values are still quantized. Still, it's an interesting use of flash technology. I guess it also means the data can still be backed up and transmitted digitally.

When I was in the Navy, the fire control systems where analog; they used syncros and servos to compute in real-time the solution to fire projectiles onto targets.

Great video. Nice to hear more on analog chips and how they are important. Being a shareholder in Spectra7 active copper cable running on 800G analog chips, this explains a great deal how it’s used. Keep it up

I love vinyl records, since I started playing my dad's collection at the tender age of 7 or 8 in the early 80s. However, I fully embrace digital music as well. Being able to carry a lot of music around in my phone is wild.

When I was in college low these many decades ago, we used individually interconnected analog boxes to solve differential equations…

At Oregon State University in the 60s we started by wiring logic boards to do tasks to a deck of punched cards. We later moved to an IBM1620 that was a mix of analog (had dials to set for many engineering functions) and digital (could run an early version of Fortran II). Used tons of power compared to the CDC 3300 that replaced it. Now that digital power requirements are dropping for both CPUs and memory I can see Analog Computers used for a specific tasks but not for General Purpose Computing. One of the issues with Analog Computing is numbers like Pi or e out to multiple decimal points required for many scientific calculations and transforms like Fourier and LaPlas.

Analog computing definitely has a place. I don't think it will change mainstream computing, but certainly for application specific uses as you described, it could be a game changer. Data storage seems to be another beneficial area since currently we would need to use 8 bytes per floating point number. With an analog solution a single analog data point could represent that number.

I listen to plenty of digital music all the time: while Im working, driving, doing stuff around the house or gaming. However, when I want to sit down and dedicate my time to listening to an album I much prefer the experience of playing vinyl records. The entire ritual of removing the record from the sleeve, placing it on the platter, brushing it and dropping the needle then settling in to look over the big, beautiful presentation of the album art and liner notes while I listen and enjoy a coffee or whiskey is extremely satisfying and relaxing, almost therapeutic in a meditative way. While there are arguments about sound quality that can be made to support either digital or vinyl, at the end of the day it's the unique experience of engaging with vinyl as a medium that made me, and has kept me, a fan of vinyl records.