Hmm. Interesting. Instead of having the machine look at people's designs, why not have all the designers put into words, what their knowledge is, and their techniques of layout, so it can be combined into software steps and rules? That way the software would be at least as good as a human. Interview all of them. Maybe that would help. Also physical modeling, and quantum computers maybe? It shouldn't cost much runtime for one chip. That's an interesting problem. I thought all chip design was automated now. Cooll.
In the mid 80's I worked at a silicon design firm. Mostly 3 micron gate arrays. My boss and a really sharp engineer from the client company designed a modem chip. Their first silicon worked fine which elevated them past hero status to God of silicon design status. Mixed signal chip design is something like witchcraft. Thanks for another great video!
@@tomaszapata I was trained in basic electronics by the US military and I had a computer programming class at the community college. On the job training from tech to engineer.
Ok, let's try to identify some of these images for fun. 0:06 A DC/DC converter module of some kind. Transformer and optocoupler in the middle of the board to isolate the primary (right) form the secondary (left). Could be for use in telco for converting 48 V to a local voltage like 5 V or 3.3 V for example. 0:13 Some analog chip produced by PMI (Precision Monolithics). Could for example be an opamp. 0:31 The title "analog delay" combined with delay stages taking opposite clock signals φ_1 and φ_2 points toward a BBD (bucket-brigade delay). This is an analog device delaying an analog signal by feeding it through a series of capacitors which each holds a sample temporarily before passing it on. I don't recognize the circuit. I would've loved for this to be either Serge's Wilson Analog Delay or Buchla's model 277 (modular synth devices) but it doesn't look like it. 2:09 As the title suggests, an analog multiplier circuit. What's neat about this circuit is that it's actually a logarithmic converters followed by a summer followed by an exponential converter. In other words, it turns a multiplication into an addition according to the laws of logarithms and exponentials: a*b=e^(log(a)+log(b)) 2:51 Looks like the same analog delay circuit as in the title card. Through the magic of reverse image search, this turns out to be an early 1976 guitar pedal prototype circuit which inspired MXR delay pedals. 3:05 The title cards for two previous videos. To the left, the board for a HackRF One, a SDR (software defined radio) which can be used to receive various radio signals. To the right, something that looks very much like a computer motherboard, with PCI or RAM slots. 4:27 Looks like a RF demodulator going into an AGC (automatic gain control) then into an amplifier feeding a builtin speaker. Ie a radio receiver. The negative voltage supply (positive ground) combined with liberal use of transformers, suggests a vintage circuit. 5:00 LM317 internal diagram. LM317 is a classic adjustable, linear voltage regulator. The schematic of the internal diagram is drawn in the datasheet. 7:18 Looks like a circuit for exercising your understanding of Kirchhoff's Law in school. Fun stuff. 10:52 A delicious cake, which I'm informed may be a lie. 12:30 Looks like a sawtooth wave generator. The second opamp is setup as a comparator, which will trigger the MOSFET and reset the phase when the voltage goes above 2.5 V, plus a little bit of positive feedback to encourage the opamp a bit to work as a comparator, which it's otherwise reluctant to do. Also, screenshots are hard. 12:44 Not a schematic, but just a note that it's difficult for AI to generate grids apparently. The lines are wobbly and the lines never seem to line up on both sides when they're obscured by some object. 15:18 Interesting. I wonder what this is. Two CPU/FPGA/whatever devices hooked up to memory, by the looks of it. And then IDC headers. This is probably some industrial or test equipment, like the digital portion of an oscilloscope.
@@bigmedge nah, it's an all in one radio. Lower left is the tuner, with the dashed line being the coupling to the antenna, and the adjustable capacitor in parallel with an inductor being used to tune the frequency. On the right side you see the amplifier section with a 1 transistor preamp and two transistor power amp, followed by what looks like a built-in speaker. Lots of transformers is just how they built stuff back then because of habit from the tube days, and the price of transistors. Transistors were cutting edge technology, and transformers were something you could manufacture easily with cheap labor. If you could solve a problem without adding more transistors, that's what you did. Today it's the opposite. Transistors, especially in integrated circuits, can be mass produced for almost nothing in a highly automated factory but transformers still require (now more expensive) manual labor.
The complicity of analog design is worse when you include temperature sensitivities in the transistor and material performance. My old company (name withheld) included a Power On Rest circuit that gated power to internal logic so that it always started up in a known state. The circuit was updated in a routine design revision that passed all qualification tests across temperature and worked great in industrial grade applications, but came to a screaching halt when it was sold as an automotive product. Process variation of some lots allowed the temperature response of the POR to shift, locking the chip up when you turned it on in a narrow temperature window, A minor issue for a cell phone, but pretty painful if you are trying to unload VW's from a ship in the winter (the ECU wouldn't function). As you noted, analog design is an art, one where you have to know how all the paints mix and interact. Nice job on the presentation, I hope your father was able to retire happily.
Underrated comment. Temperature sensitivity affects analog circuits in so many ways, and that is partly why certain designs performed better as an IC than as discrete components.
Quite true. Temperature sensitivity is just sort of part of the analog design flow. It's even part of why there's special analog processes - analog processes often have multiple types of resistors with different TCs and sheet rhos - finding a balance between the right TC, area, and performance means you have to pick the right one for the application and maybe make tradeoffs. And it's not just that. We often do things like design amplifiers with current biases that change over temperature so that the performance of the amplifier stays relatively the same over temperature. And it goes on and on. Another place where the art shows through is on stuff like DC-DC converters. You have, on the same piece of silicon, incredibly noisy power switches doing many amps of current and delicate low noise references, amplifiers, and ADCs and DACs. Making it so those two things can coexist with good separation is quite an art indeed.
@@brodriguez11000 Kinda is. I'm sure at some point AI will come in and help with design, but it's going to be a long time before it's a replacement for a human. Even in other fields, like programming, AI is a really useful tool, but it still very much needs a human to guide it, direct it, critique it, debug it, etc. Now ultimately, that may mean the human is thinking about stuff like top level architecture and requirements while the AI is doing low level work, but I don't see humans not being part of the loop for quite some time.
always reminds me of my own immigrant dad, who sacrificed everything he had, without knowing if he will ever be able to achieve his dream of being a physician, to be able to raise his son outside of a warzone. He himself said he wouldn’t have been able to do any of that without my mother, and i would dare to doubt it for a second. I’ll never stop being grateful for what they did for me, I’m looking forward to being able to honor them by being a good dad myself one day.
As an analog/RF IC designer I can relate. About a dozen years ago when I was working at AMD our CAD group made an attempt to automate analog design to redesign chips shrinking to a smaller process node. It did this by interpolating the I-V curves to resize transistors. The results were laughably amusing and none of my IC design peers ever lost sleep worrying about design automation. Layout for analog circuits might be an even tougher nut to crack because an alogrithm would have to iterate on extracted simulation, layout, LVS/DRC checks, re-extraction, and back to simulation. Given the size of extracted netlists, I would expected the required hardware compute resources to require a supercomputer. LOL. But all of that was before AI. Fortunately, I'm close to retirement now so I expect to skate by just as your father did. Not so sure about the next generation of circuit designers, however ....
My perspective of 50 years of working in electronic industry is digital is a very narrow subset of analog. Resistance, and/or, impedance plus capacitance, and the one thing you left out in your talk is inductance. But you are correct and that it takes a few people with the artistry to do a good job in design.
As an early career analog designer, I have to say that you generally got the design flow right, except that nowadays almost no one is using a paper and pencil to do anything significant. Usually things will be sized and played with in a simulator, experienced designers "have a feeling" about what sizing is going to work, and are usually right. The complexities of BSIM-BULK models (formerly BSIM6) means that doing anything on paper is fruitless, though there is some nice guiding principles in the EKV model. Reductions in supply rail headroom have also meant we have to do most things in weak inversion now, and that changes your current characteristic from quadratic to exponential, another weirdness that is different from what is taught in university. AI training really seems hard to achieve, though probably not impossible considering what LLMs are doing these days. But one of the key problems is that there are many criterion for "good" and so in order for a computer to optimise, this optimisation has to be set up in the outset. This means that two different analog designers might have very different goals with an assistant AI, one might care more about their area specification, another might be more focused on how they are going to achieve their dynamic range spec with the power consumption spec. As you said, it's a very high-dimensional space, I'm not sure how one AI tool can really guide you through all of that. That's to say nothing on how you are attaining this dataset to train on, where's that coming from? Also, who is qualified to do the annotation of data for a dataset? Lots of hard problems to solve in this domain.
13:02 AlphaGo was not actually trained on huge datasets. The reason it is capable of beating the world's best players with *novel* techniques is because it was not trained by watching humans play. Instead, it played against itself and gradually learned over millions (or more?) iterations while grading its performance.
I used to be an analog and mixed mode ic designer. I had LOTS of fun. I especially enjoyed the challenges of simulating the circuits properly. And what a feeling when the first silicon arrived on the bench and it worked exactly like in the simulator! Of course sometimes certain behaviors of the actual chip remained a mistery...
@@manonamission2000 well, it's been over 20 years ago. If I remember correctly I had an offset problem: the measurement and the involved math showed beyond any doubt that there was a constant offset causing a systematic error in the slope of a ramp. But my colleagues and I looked everywhere (circuit schematic, layout, design parameters) but everything seemd right. In the last run we even chose a more expensive mask set, but the offset was still there. Fortunately the parameter was not a fundamental one, so we simply relaxed the spec and the device could enter production. But I could never figure out the root cause.
Jon - You don't have to eat the full slice of black forest cake. The real world is analog - you can have have a 1/2 slice... 1/4 slice... or just grab the cherry from the top. As always, a great video on a very complex topic.
As an electronics repair tech for a large US manufacturer in the UK, I work on and use precision analogue electrical measuring instruments every day. I almost never gave any thought to the complex analogue designs inside the many many ICs used in this equipment. Being a designer of the circuitry in this high accuracy analogue equipment is a huge challenge in itself. The design engineers job is having to take into account parasitic capacitances, inductances and resistances. Having made a few analogue designs from discrete and IC components, I know how challenging some of these can be. I sometimes think it is all magic, and I have been in the industry since I left college in 1984. I take my hat off to those who design analogue circuitry on a microscopic level, and am glad to see you honour your father in this piece. Can I also just add that the sweet baked confection shown is a gateau, not a cake. Gateaus tend to have more than one layer of filling, usually a combination of buttercream icing and a fruit jam or compote. The main construction is formed of a light sponge. Son of an electrician and a baker.....
Same :) I'm still caught up between doing analog which I love more, or digital, which, considering _most_ of my classmates and seniors in VLSI are doing, seems a more 'charted' territory
Nice analysis. In my view, we can further divide the field into 'needs the best performance' and 'works well enough'. I think for now analog designers have job security for the high performance area. However, I believe that we'll see the growth of automated and AI tools for 'good enough' analog. Especially in the open source field, as it's a lot cheaper to test and develop. One of the criticisms of open source ASIC tools I've heard is that the older process sizes are not relevant, but in fact they are still very highly regarded by analog engineers for the reasons you outlined in your video. Additional interesting reading can be found here: "Open-Source and Autonomous Temperature Sensor Generator Verified With 64 Instances in SkyWater 130 nm for Comprehensive Design Space Exploration"
There's actually quite a few other reasons why newer processes aren't well regarded by analog designers. First, those transistors are often optimized for good digital performance without any regard to analog performance. This means they might have things like a poor gm curve. Or lots of equivalent output resistance. Or bad subthreshold leakage. Or poor matching characteristics. Or just plain weird analog behavior (like gate oxide tunneling). Or... Second, those processes are rarely setup for good analog circuits. So, they will lack things like the ability to mix and match transistors with different threshold voltages. Or perhaps lack good resistor options, particularly high sheet resistance or low temperature coefficient resistors. And stuff like BJTs (yes, still common on analog designs, especially references) is virtually always an afterthought on those processes. It's usually some parasitic device with stuff like really bad betas (current gain from base to collector) and lots of undesirable parasitic devices and leakages. And zener or schottkey diodes are virtually nonexistent on those nodes. They will also rarely have things like extended drain devices for high voltage operation. It's also often really helpful to be able to put devices in different wells, both for isolation purposes, but also so you can float transistors above (and sometimes below) the substrate and that ability will be limited on a digital process or come with major drawbacks. Third, alot of analog circuits go into power converters or RF. On both of those, it's rarely the active analog circuitry that dictates most of the die area. For power converters, it's usually the power switches, which will dominate the die area and don't benefit much from ultra small geometry stuff. For RF circuits, it's often stuff like inductors and transformers that dominate the chip. The actual analog circuitry is typically tiny, then you usually have some moderately sized mixed signal stuff to digitize the analog, then a big digital block to process it all (often on a different die - usually on a process optimized for digital). A good example would to pull up a die shot of the ESP8266 (a low end microcontroller with integrated WiFi). Zeptobars has a nice one. The digital is actually a pretty big chunk, but the single most prominent features are the big inductors and transformers for the RF (usually large octagonal or circular constructs). Fourth, most of the actual analog processes get less attention and less money. So they are often old digital processes that have had analog stuff bolted on after the fact. Not only does that make them less than ideal for analog, but it all but guarantees it will be an old process. Which is usually fine because the analog rarely actually needs stuff like finer geometry (doesn't actually make the chip that much smaller), so why pay for more expensive masks and processing? And why incur the MUCH more difficult design rules that come with fine geometries?
Wow.. that's a bit creepy.. How does it feel to be a time traveler @matthewvenn ? Is the trip backwards through time (and presumably forward through time.. if you're reading this) as nausea inducing as I imagine it to be? Or is it like surfing a gentle wave for a time traveling bad@$$ such as yourself?
Analog circuit design using discrete devices is an art by itself. Placing an analog circuit onto an integrated circuit is art at the grandmaster level.
As someone who has *tried* to use the Berkley Analog Generator (BAG) one of the main problems is setting the system up. First it is a python library (built by grad students) with lot's of dependencies, these are not always readily available on the operating systems used to run EDA software (*cough* RHEL 7 *cough*). After that hurdle is cleared, BAG needs a set of base layouts that the software can array copy (for lack of better words) next to each other. These base layouts are necessary to build up larger transistors with the flexibility to meet the design needs. These layouts are difficult to make due to the complicated design rules for current-gen and leading edge processes. Also, the work to build the layouts cannot be shared between processes as they are, in general, quite different from each other. All this to say, BAG has significant adoption costs associated with it.
Just to clarify, there's no metal in the center of the FinFET, that's silicon--an even worse conductor (semi-conductor in fact), which means even more resistance.
20 years ago somebody brought me on as a contractor to add digital control to an otherwise analog system. There was zero chance to make it work as they had not bothered to properly characterize their design and understand the component constraints needed for reliability and EMI/RFI compatibility & stability. 🤦🏻
Working at an analog company we only considered ESD protection the black magic. Most analog design was fairly automated after proper modeling. Qualifying a new process was very in depth, and there were a few gotcha’s that popped up in the first designs with nearby devices or interconnects but those could be worked in design rules and coded into automated design tools. We would occasionally find a new problem in mature processes, but those learnings were quickly tested for in other processes and proactively prevented. We weren’t at 7nm, or 32nm, but the analog circuits were always larger geometry then the digital portions even in analog heavy designs.
Put thermal accurate thermal models behind each component and a spice of minority model of the substrate connection it turns to less an art but more engineering.
1:00 It is no longer limited to "comes from real world" QPSK, QAM are extensively used in real digital world to transfer data more than 1 bit / cycle - >10 bits per cycle. Not to mention SDR radios etc.
I worked as an FPGA digital designer with an office next to group that did mixed-signal chip design for imaging readout chips. This was a little different from some of the examples in the video because the imaging chips were and are much more analog than digital, and it was no secret that these mostly analog chip designers were the technical stars of the company. I had the privilege to be asked to sit in on one design review and was mightily impressed with the thought and skill that shown through in that presentation and discussion.
@@homelessrobotNo, they really aren't all digital. Up to 20% of each digital clock cycle of many architectures is spent in the linear range of the transistors (this is a large part of why they draw power, and why higher clock speeds draw geometrically more power) and not in saturation. Transmission lines, may they be data busses, networking cables, or power distribution wiring are all also inherently analog.
@@DrewNorthup what I was refering to is the planck constant. The energy of particles being in discrete increment, and time passing in discrete windows within which no ordering or subsequence of smaller moments can be discerned.
Great stuff as always. There's been a change of tone compared to your earlier work which was more academic. I suppose I prefer that, but I am glad you are having more fun with these! Please keep this stuff going.
Hi. I am engineer. Analogue design will be completely automated, if not already in some heavy projects, because the design reuse has reached its maturity early 2000, i.e. now there is only the need for advanced heuristics and optimization algorithms to achieve a middle range complex design of about 10000 transistors (we are not speaking of digital designs since these have pretty fixed rules(Lynn Conway anyone?). Unfortunately, the problem is that the new generation is not interested in analogue design, as being confident that digital systems will prevail or even they can outsource their analogue designs in the East. A bravo to Asianometry for this great presentation.
I worked for years in the EDA SW team at HP/Agilent working with Test & Measurement teams. You are not kidding, analog design, especially very high frequency analog is freaking insanely complex & difficult...
Analog design is the Gordian knot of design disciplines. AI as Alexander Invictus however will be denied a single sword to slice through it. It will have to unravel the knot every time. If it can, let everyone beware.
My first analog circuit was a DC/DC converter with multiple inputs, single output to define the desired output voltage given the multiple inputs (it had power delivery limit, desired speed and temperature), it was designed to move a DC motor for a prototype car. Why did I made it analog? I dunno, I just loved the idea of making something different using something no one around me was able to even try to copy
Touching, funny, deep! Anyone with even a passing familiarity with circuit design can appreciate this! Thank you for brining this human element to a highly technical subject! Kudos and godspeed!
As someone working in IC design, I congratulate you for the accuracy and clarity of the video! I'd just make a very minor correction: verification starts before, and is done in parallel with, the physical design. It only finishes after it, because it needs delay back annotation .
As newbie analog layout engineer (not circuit design) the main time-consuming part for me with a new pdk is studying the DRM and making sure all the rule checks/requirements from circuit designer are cleared after the design and of course all that PEX/EM/IR/ERC issues and what not are met. My job sometimes feels like laying bricks following specific rules after the main designer (circuit designer) has designed the stuff for the building. I think the layout part could get automated at some time in the future (in fact some things are already done using scripts that does not need human intervention). Also, fun fact: Major companies usually outsource the manual jobs to 3rd world countries like India/Bangladesh. However, apple still maintains a very strict policy and their contractors have to be (I believe) physically present in their country/office. No wonder apple still rules the chip market with their efficient hardware-software combination, and their chips are so optimized whereas others are 'if it works it works'.
I only randomly view your videos, but it was cool to hear that your dad designed analog chips and that you were able to explore his realm with this video.
Yep. We had implemented everything in analog manner before this digital technology era. It was extremely hard if an implementation had to handle very high frequency signal. So, I feel that the analog tv systems were a miracle now a days. Anyway, thanks for your great shows.
Those who aspires to be analog designer should have many first-hand experiences on discrete high frequency analog circuits design first in early age. You can experience many varied and weird parasitic behavior of analog circuits on such RF circuits. Familiarize yourself with many historic analog circuits design like radio, TV, classic amp, radio transmitter, op amp circuits. Study workings of the circuit carefully down to a resistor. Once you are accustomed and learned such strange and intricate nature of analog circuits, then study carefully circuit, layout, and floor plan of famous classic analog integrated circuits like LM741, NE555, LM324. etc. Compare SPICE simulation result and real circuit behavior. Know the difference of SPICE mode and real world integrated circuits. Try to understand why they are designed circuit and layout such way. You will get lots of insights from such classic designs.
@@stevebabiak6997 Actually it is good idea. Classic vacuum tube amplifier circuit has quite well-designed linear circuit. FET (not bipolar BJT) works like vacuum tube. So, you can learn and borrow many ideas for MOSFET circuits from early vacuum tube circuits. 5 tubes superheterodyne vacuum AM radio receiver is good example to study to learn about analog electronics. I myself leaned from classic vacuum amp and radio.
@@youcantata - correct, FET and vacuum tubes have some similarities. But there are cool devices too, like tuning eyes for example - that don’t have a direct analog in the semiconductor world.
It might sound weird, but capacitors are analog. And tiny little capacitors make up RAM. We try and clear them, and recharge them, within tolerances, but it's definitely not a binary value between those two charge values.
All devices are analog. It is only the interpretation that we put on the signals that is digital. But whether a circuit's signals are destined for analog or digital interpretation places very different constraints on its design.
Memory cell uses cross coupled devices tied into a positive feedback loop. To write a memory cell, you need a charge pump & memory driver(both are analog circuits).
As @danmenes3143 says, everything is essentially an analog device, simply because it exists in the real world. We just truncate that function to digital, for ease-of-use and manufacturing. I wouldn't be surprised if we end up with pseudo-super-conductors, where it's not lossless, just really fast and efficient, after more research into the art, science, and craft of analog circuits and devices.
This is true for all electronics. "Digital" electronics just means they work close enough to an ideal digital circuit that we can assume it's digital. This is not always the case though, case-in-point for your example of DRAM, the Rowhammer attack, which repeatedly wrote to strategically chosen memory locations to influence other locations in memory.
Nice coverage of the topic! The core of the talk was about analog layout automation, yet there are many analog design automation methods (including ML based approaches) worth mentioning. Your channel has some of the best videos ever covering semiconductors technologies/manufacturing. It's great to know that all falls back to your father being an analog designer!
Yay! I love analog circuit design, though strictly as a hobbyist. I have discovered areas where analog circuits still perform better than their digital equivalents, for example fast PID controllers. The closing sentence sums up well my feelings about it being art.
Woah, it never occured to me that physical designers for digital logic lost jobs, I thought most of them just switched to working the CADs instead of doing that by hand.
Analog chip design is very tough. This is a great video that can be touched on multiple times. Safety critical devices pretty much only take analog inputs and as we innovate some of the new boards are not as good as the old ones from the 80's !!! We moved backwards in analog chip design !!
You left out power supplies, which are a major block of analog on soc parts. Analog capacitance and magnetics are physically big, so they take up a lot of space as do the deep wells. Rf design such as wifi cellular and Bluetooth also have considerable analog preprocessors, but are part of most soc's. The topology of analog can switch from unbalanced to balanced in integration, this driving up complexity. In the end, I don't hold out much hope for ai, but a combination of ai with quantum computing that can model multi dimensional parasitics, now we are talking.
1) AI is trained on data from the Internet. 2) AI outputs data to the Internet. 3) Goto 1 ... Exponentially growing AI-feedback ... you now see how this ends :(
A similar thing that has always baffled me is analog mechanical computers like the fire control system in Iowa battleships. Which as far as I know remained a lot more accurate in it's firing solutions than many digital types that followed. And since it mixed inputs continually it also gave you solutions real time. The only not real time part was the human operator needed to feed the wind speed or some of the other inputs into the system. But other than that it was solving an analog equation with the given inputs.. mechanically. As a student of the binary age I've always thought that was really crazy that someone thought of doing things like that.
Yea, I watch Ryan over @BattlshipNewJersey as well. Digital is like quantization - it's discrete. Analog natively supports 1/2, 1/4, 1/8... 1/n.. 1/n*2... increments. Terms like a "smidge" are perfectly valid. Also, *all* of the inputs on the Mk 38 Range keeper along with the Mk 41 stable vertical are analog. Distance, bearing, temperature, humidity, on and on. Analog calculating analog is ideal in this situation. The inputs are the impetus for the outputs. The output changes continuously with the inputs. The firing solution is a natural result of the inputs. It was 40 years of development before the analog to digital encoders, and the digital computers were fast and reliable enough for the Navy to make the transition. Even then, the firing solution was discrete. Inputs became fixed so a firing solution could be calculated and made available when the trigger is pulled on the pineapple.
I started my analog design career with a CK722 transistor which I promptly burned out. I did a lot of analog designs using analog chips like the 741 op amp. I got into digital electronics before TTL came out. Went on to use the first microprocessors but they were too slow to do much of anything. I invented a digital circuit design plan that I called 'random logic' which implemented software using hardware. My random logic was much faster than any CPU could do.
Just like any truly creative art. AI is a tool to be used. It's really handy. It (for now) cannot replace the nuances of human. AI can conjure unexpected things that might work with a little tweaking from someone with experience. Saves a lot of trial and error. Doesn't replace the human. Human can do more in less time. Positive feedback loop. Remember the bad video transitions of the 80's and early 90's? Yeah. That will happen and then people will hate it, stop buying it and it will go away. Then it will make a come back for a while and then be regarded as lazy again. The real art is timeless.
Love your channel big bro. I used to be more of a software type of guy, but you kinda nudged me into messing around with hardware alot lol. I might just get an FPGA and see if I can make my own stuff in the near future lol
Today, they employ various building blocks to create larger groups or modules. These building blocks include oscillators, switches, amplifiers, and so on, which are assembled into functional units. However, the approach taken depends on the specific RF (Radio Frequency) components they are working with. When dealing with GHz (Gigahertz) parts, the design process often starts from scratch, with components being integrated into a complete system. In contrast, for low-frequency parts, the approach typically involves assembling pre-existing modules and grouping them together to form the final system.
Something not mentioned in the circuit layout part is the inability for algorithms to route circuits in a determinable amount of time. Most algos used in programs have a best/worst/average case (BigO notation) that is polynomial which mean time to complete is an exponent of the items to iterate over. Some algos are deterministic but some, like the travelling salesman problem, are nondeterministic and thus cannot predictably be solved in the most efficient way possible by a machine.
If you can create a Go champion with AI, you can do the same with analog circuit design. It's "just" that nobody's thrown the money and effort into it (yet).
nice :) I really struggle with the analogue parts of my projects :) I notice quite a movement to re-create old analogue chips like sound chips with FPGA emulation based on digital filters and some very creative bitstream shenanigans. That is open to AI approach too I think ? I recall some mobile robot gait generation approaches based on AI gen of bitstreams for FPGA ... with class D amps all sorts of oddness become possible on the analogue generation side.
I would like to know all about analogue sound chip FPGA bitstream shenanigans, if that's actually what they are and not just clever digital programming. You might be aware of Adrian Thompson's classic 1997 paper "An Evolved Circuit, Intrinsic in Silicon, Entwined with Physics". He applied a genetic evolution approach directly to the bitstream to evolve a tone detector that could discriminate a 1 kHz and 10 kHz tone on a FPGA with no external timing reference. It got there eventually. But the odd thing was that the circuit relied on cells that were disconnected to function. And when he tried moving the bitstream to a different of the same model, it no longer work. The genetic algorithm had evolved to live on one specific piece of hardware.
You need to interview guys like Doug Curtis, Dave Smith, Tom Oberheim, Dave Rossum... if you want to know about analogue IC design. Sadly Dave Smith passed last year but the others are still around and putting out products based on analogue circuitry.
10:20 "Shrinking metal pitch from 80 nanometers to 48 nm raises the line resistance by 6 times" 80/48 = 5/3 = 1.66 (5/3)^4 = 6.4 if this exponent continues to 10 nm -> (80/10)^4 => 4096 times larger... crazy
As always, great video. But you forgot to mention the most exiting one: parasitic inductance. It really can do some fun stuff together with parasitic capacitance 😅 And for all the people that are impressed by the diagrams shown in this video: the white space… that’s were all the parasitics are hiding!
Study in Electronics engineering almost 30 years ago but didn't work in this industry after that. This video reminds me my old day. The day I can understand and have to calculate each bloc to double check things. Talk about Spice ... Man I really miss it. Another software that comes up in my mind is MATLAB.
As a hobbyist and looking into pcb design I was hoping AI would help with analog design. Looks like I’m going to learn the hard way which is entirely much better any way because it forces you to understand the intricacies of the design and allows you to troubleshoot when needed. Cheers
Famous last words... xD, but whether they are mine or yours I do not know. I personally believe that AI is quite suited for design processes with a lot of variables, although humans will always guide the process to some extent, but on more and more abstract levels moving forward. So with less input garnering more control. I think the key is connecting an LLM to a generative AI and an automated prover. The generative AI can adversarially learn to move across the state-space given by the proving process (it mimics validity). If that proving process includes certain effects, the generative AI will be capable of handling those too. This generative AI can be initially setup to be partly driven by some huge random input value, which then can be replaced by an LLM setup to influence the design outcomes. The LLM learns to output values that cause the circuit design to mimic those it learned elsewhere. A good start would probably be all the textbooks about circuit design. This should enable high-level parsing of abstract input (and iterative guidance of the design process).
excellent video about the difficulty of analog circuits. I'm hopeful someone can train a model to make this easier. I have to think TSMC is working on this problem internally.
As someone struggling and constantly befuddled by simple circuit designs... A lot things clicked watching this. 10/10 quality content and knowledge on this channel!
Ah,yes, design a Butterworth filter for the following frequency range, with a Q TBD. Excellent overview Jon. But don't let my review go to your head, please.
ASML / TSMC probably need to replace the copper with silver/ platinum/ gold or alloys thereof, for the sub 3nm 1nm & so on 0.6nm nodes since copper parasitic resistance in shrinking pitch sizes increases so rapidly as a function of the cross section decreasing.
Can you elaborate on the issue with the proprietary PDKs? Looking at the GitHub for both align and magical they appear to be BSD licensed, so there shouldn’t be any issues there, do the PDKs have some kind of anti-OSS licensing associated with them?
Analog circuit design is an art, if good engineers designs it with understanding, of what the circuit has to do and later tests and layouts it. If neural networks do it, you likely get a mess, that somehow works, but with no understanding in how and why.
But AI by itself would just put out garbage. Analog Design encompasses a wide range of fields: PLL to AFE to ADC to switching regulators to RF design. Too broad for AI.
I'm a senior engineer. I also see analog design as an art that needs time to master, which is being lost in a world were something that has more than 5 years is too old. Thank you for your kind words.
I would love if you made a video on the Semiconductor supply industry. like who makes the machines that make the parts for the semiconductor industry. i work for a CNC dealership. Matsuura is a great company and they have this really cool metal 3D printer that also mills at the same time
@asianometry Good to know that me as real world useless magister of philosophy came to the very same conclusion as a technical expert like you = "Life is analog". Exactly one year ago as a notorious coffee loiterer and registered inventory I ve experienced transcendental rational blackout and stroke of enlightenment of my own ingenuity and I opened my a own coffee house (bar) called Barista (what a original, doesn't it) with main claim: Barista - "The analog type of coffee house" and often added with subtitle: Real life is analog... So in conclusion. Despite: a) I am totally out of target group, b) knowing about semiconductor chip design less then amount of milk in ristretto, and... c) main goal of this video is to create ideal circumstances for tech savvy monsters to cause them at least week long chain of wet dreams... Despite all af this, I really like this video :-)
My first thought in this video, was correct selection from about millions of existing semiconductors in our world in designing analogue circuits boards. AI could help a lot in selecting these in function and low costs. This could speed up designing processes. // Would you do the same video in analogue design of elemental digital circuits components? => Design rules of transistors, gates, flip-flops, synapsis, neuronal circuits, ... ? Thank you for this video.
analog designing is to ai driving a car versus to a human, ai can only work on syntax but human can work on both syntax and semantics. in less nerd terms, ai knows the best route to take but humans will take a route because he needs to.
This episode oozes all the things that makes Asianometry great. Expert, deep, quirky, silly, and interconnected.Thank you for what you do!
I'm mainly hear for the chocolate cake news.
@@guaposneeze Says the clown that doesn't know the difference between hear, & here...
Yup
Hmm. Interesting. Instead of having the machine look at people's designs, why not have all the designers put into words, what their knowledge is, and their techniques of layout, so it can be combined into software steps and rules? That way the software would be at least as good as a human. Interview all of them. Maybe that would help. Also physical modeling, and quantum computers maybe? It shouldn't cost much runtime for one chip. That's an interesting problem. I thought all chip design was automated now. Cooll.
@@Sum_Tings_Wong Calm down a bit. Your response reads as if you were personally attacked by a misspelling
In the mid 80's I worked at a silicon design firm. Mostly 3 micron gate arrays. My boss and a really sharp engineer from the client company designed a modem chip. Their first silicon worked fine which elevated them past hero status to God of silicon design status. Mixed signal chip design is something like witchcraft.
Thanks for another great video!
and?
The *first* silicon worked, he said. That's pretty rare.@@kallucelfrumos4946
What did you study @jonpattison?
@@tomaszapata I was trained in basic electronics by the US military and I had a computer programming class at the community college. On the job training from tech to engineer.
Ok, let's try to identify some of these images for fun.
0:06 A DC/DC converter module of some kind. Transformer and optocoupler in the middle of the board to isolate the primary (right) form the secondary (left). Could be for use in telco for converting 48 V to a local voltage like 5 V or 3.3 V for example.
0:13 Some analog chip produced by PMI (Precision Monolithics). Could for example be an opamp.
0:31 The title "analog delay" combined with delay stages taking opposite clock signals φ_1 and φ_2 points toward a BBD (bucket-brigade delay). This is an analog device delaying an analog signal by feeding it through a series of capacitors which each holds a sample temporarily before passing it on. I don't recognize the circuit. I would've loved for this to be either Serge's Wilson Analog Delay or Buchla's model 277 (modular synth devices) but it doesn't look like it.
2:09 As the title suggests, an analog multiplier circuit. What's neat about this circuit is that it's actually a logarithmic converters followed by a summer followed by an exponential converter. In other words, it turns a multiplication into an addition according to the laws of logarithms and exponentials: a*b=e^(log(a)+log(b))
2:51 Looks like the same analog delay circuit as in the title card. Through the magic of reverse image search, this turns out to be an early 1976 guitar pedal prototype circuit which inspired MXR delay pedals.
3:05 The title cards for two previous videos. To the left, the board for a HackRF One, a SDR (software defined radio) which can be used to receive various radio signals. To the right, something that looks very much like a computer motherboard, with PCI or RAM slots.
4:27 Looks like a RF demodulator going into an AGC (automatic gain control) then into an amplifier feeding a builtin speaker. Ie a radio receiver. The negative voltage supply (positive ground) combined with liberal use of transformers, suggests a vintage circuit.
5:00 LM317 internal diagram. LM317 is a classic adjustable, linear voltage regulator. The schematic of the internal diagram is drawn in the datasheet.
7:18 Looks like a circuit for exercising your understanding of Kirchhoff's Law in school. Fun stuff.
10:52 A delicious cake, which I'm informed may be a lie.
12:30 Looks like a sawtooth wave generator. The second opamp is setup as a comparator, which will trigger the MOSFET and reset the phase when the voltage goes above 2.5 V, plus a little bit of positive feedback to encourage the opamp a bit to work as a comparator, which it's otherwise reluctant to do. Also, screenshots are hard.
12:44 Not a schematic, but just a note that it's difficult for AI to generate grids apparently. The lines are wobbly and the lines never seem to line up on both sides when they're obscured by some object.
15:18 Interesting. I wonder what this is. Two CPU/FPGA/whatever devices hooked up to memory, by the looks of it. And then IDC headers. This is probably some industrial or test equipment, like the digital portion of an oscilloscope.
Thanks for sharing the insight! 😃
Nice to know that AI struggle to draw grids.
I know old school audio amplifiers had a ton transformers in them , so perhaps the schematic is for 1 of the amp’s stages
@@bigmedge nah, it's an all in one radio. Lower left is the tuner, with the dashed line being the coupling to the antenna, and the adjustable capacitor in parallel with an inductor being used to tune the frequency. On the right side you see the amplifier section with a 1 transistor preamp and two transistor power amp, followed by what looks like a built-in speaker.
Lots of transformers is just how they built stuff back then because of habit from the tube days, and the price of transistors. Transistors were cutting edge technology, and transformers were something you could manufacture easily with cheap labor. If you could solve a problem without adding more transistors, that's what you did. Today it's the opposite. Transistors, especially in integrated circuits, can be mass produced for almost nothing in a highly automated factory but transformers still require (now more expensive) manual labor.
@@Gameboygenius
Look like an old transistorized AM radio to me, I can almost see the wax they put on the variable inductors.
The complicity of analog design is worse when you include temperature sensitivities in the transistor and material performance. My old company (name withheld) included a Power On Rest circuit that gated power to internal logic so that it always started up in a known state. The circuit was updated in a routine design revision that passed all qualification tests across temperature and worked great in industrial grade applications, but came to a screaching halt when it was sold as an automotive product. Process variation of some lots allowed the temperature response of the POR to shift, locking the chip up when you turned it on in a narrow temperature window, A minor issue for a cell phone, but pretty painful if you are trying to unload VW's from a ship in the winter (the ECU wouldn't function). As you noted, analog design is an art, one where you have to know how all the paints mix and interact. Nice job on the presentation, I hope your father was able to retire happily.
Underrated comment. Temperature sensitivity affects analog circuits in so many ways, and that is partly why certain designs performed better as an IC than as discrete components.
get better circuit design bro
Quite true. Temperature sensitivity is just sort of part of the analog design flow. It's even part of why there's special analog processes - analog processes often have multiple types of resistors with different TCs and sheet rhos - finding a balance between the right TC, area, and performance means you have to pick the right one for the application and maybe make tradeoffs. And it's not just that. We often do things like design amplifiers with current biases that change over temperature so that the performance of the amplifier stays relatively the same over temperature. And it goes on and on.
Another place where the art shows through is on stuff like DC-DC converters. You have, on the same piece of silicon, incredibly noisy power switches doing many amps of current and delicate low noise references, amplifiers, and ADCs and DACs. Making it so those two things can coexist with good separation is quite an art indeed.
@@ccoder4953 Sounds like job security.
@@brodriguez11000 Kinda is. I'm sure at some point AI will come in and help with design, but it's going to be a long time before it's a replacement for a human. Even in other fields, like programming, AI is a really useful tool, but it still very much needs a human to guide it, direct it, critique it, debug it, etc. Now ultimately, that may mean the human is thinking about stuff like top level architecture and requirements while the AI is doing low level work, but I don't see humans not being part of the loop for quite some time.
The honor you show your parents is so wonderful
You can tell he really respects his dad and what he did; that's awesome to see nowadays.
always reminds me of my own immigrant dad, who sacrificed everything he had, without knowing if he will ever be able to achieve his dream of being a physician, to be able to raise his son outside of a warzone. He himself said he wouldn’t have been able to do any of that without my mother, and i would dare to doubt it for a second. I’ll never stop being grateful for what they did for me, I’m looking forward to being able to honor them by being a good dad myself one day.
As an analog/RF IC designer I can relate. About a dozen years ago when I was working at AMD our CAD group made an attempt to automate analog design to redesign chips shrinking to a smaller process node. It did this by interpolating the I-V curves to resize transistors. The results were laughably amusing and none of my IC design peers ever lost sleep worrying about design automation. Layout for analog circuits might be an even tougher nut to crack because an alogrithm would have to iterate on extracted simulation, layout, LVS/DRC checks, re-extraction, and back to simulation. Given the size of extracted netlists, I would expected the required hardware compute resources to require a supercomputer. LOL. But all of that was before AI. Fortunately, I'm close to retirement now so I expect to skate by just as your father did. Not so sure about the next generation of circuit designers, however ....
Maybe an analog computer designing analog circuits.
All the physical characteristics of the devices change. They sure knew CAD...and only CAD.
My perspective of 50 years of working in electronic industry is digital is a very narrow subset of analog. Resistance, and/or, impedance plus capacitance, and the one thing you left out in your talk is inductance. But you are correct and that it takes a few people with the artistry to do a good job in design.
As an early career analog designer, I have to say that you generally got the design flow right, except that nowadays almost no one is using a paper and pencil to do anything significant. Usually things will be sized and played with in a simulator, experienced designers "have a feeling" about what sizing is going to work, and are usually right. The complexities of BSIM-BULK models (formerly BSIM6) means that doing anything on paper is fruitless, though there is some nice guiding principles in the EKV model. Reductions in supply rail headroom have also meant we have to do most things in weak inversion now, and that changes your current characteristic from quadratic to exponential, another weirdness that is different from what is taught in university.
AI training really seems hard to achieve, though probably not impossible considering what LLMs are doing these days. But one of the key problems is that there are many criterion for "good" and so in order for a computer to optimise, this optimisation has to be set up in the outset. This means that two different analog designers might have very different goals with an assistant AI, one might care more about their area specification, another might be more focused on how they are going to achieve their dynamic range spec with the power consumption spec. As you said, it's a very high-dimensional space, I'm not sure how one AI tool can really guide you through all of that. That's to say nothing on how you are attaining this dataset to train on, where's that coming from? Also, who is qualified to do the annotation of data for a dataset? Lots of hard problems to solve in this domain.
13:02 AlphaGo was not actually trained on huge datasets. The reason it is capable of beating the world's best players with *novel* techniques is because it was not trained by watching humans play. Instead, it played against itself and gradually learned over millions (or more?) iterations while grading its performance.
AlphaGo is trained on a huge dataset of expert data then self-play to improve. AlphaGo Zero is pure self-play data.
@@AediWangand between them both, did they lick the game plate clean?
@@Yj-Fjas I understand, yes; that is the “zero” part of alphago zero
@@AediWangThat's true
I used to be an analog and mixed mode ic designer. I had LOTS of fun. I especially enjoyed the challenges of simulating the circuits properly. And what a feeling when the first silicon arrived on the bench and it worked exactly like in the simulator! Of course sometimes certain behaviors of the actual chip remained a mistery...
care to share those unexpected behaviors?
that's how we learn how the universe really works
@@manonamission2000 well, it's been over 20 years ago. If I remember correctly I had an offset problem: the measurement and the involved math showed beyond any doubt that there was a constant offset causing a systematic error in the slope of a ramp. But my colleagues and I looked everywhere (circuit schematic, layout, design parameters) but everything seemd right. In the last run we even chose a more expensive mask set, but the offset was still there. Fortunately the parameter was not a fundamental one, so we simply relaxed the spec and the device could enter production. But I could never figure out the root cause.
Jon - You don't have to eat the full slice of black forest cake. The real world is analog - you can have have a 1/2 slice... 1/4 slice... or just grab the cherry from the top. As always, a great video on a very complex topic.
I'm not normally down with the goofy tangents and memes in dry educational videos but you ride the line.
I absolutely died in his Japanese Computer video when "actual footage of invasion of japanese market" which was moby dick ramming the piquant. 🤣
I think he goes over at times. Little cringy, just allow for the natural irony of the content to be the funny
he's awesome and you're jealous!!@@JBrinx18
I laugh freely at them. Keep em coming!
@@JBrinx18 No one is keeping you here against your will. If you don't like the content, watch something else.
As an electronics repair tech for a large US manufacturer in the UK, I work on and use precision analogue electrical measuring instruments every day. I almost never gave any thought to the complex analogue designs inside the many many ICs used in this equipment. Being a designer of the circuitry in this high accuracy analogue equipment is a huge challenge in itself. The design engineers job is having to take into account parasitic capacitances, inductances and resistances. Having made a few analogue designs from discrete and IC components, I know how challenging some of these can be. I sometimes think it is all magic, and I have been in the industry since I left college in 1984.
I take my hat off to those who design analogue circuitry on a microscopic level, and am glad to see you honour your father in this piece.
Can I also just add that the sweet baked confection shown is a gateau, not a cake. Gateaus tend to have more than one layer of filling, usually a combination of buttercream icing and a fruit jam or compote. The main construction is formed of a light sponge.
Son of an electrician and a baker.....
As someone who aspires to work as an analog designer this was really encouraging to hear.
It is a very nieche occupation that is taught in very few places.
@@0MoTheG
so not good?
Same :) I'm still caught up between doing analog which I love more, or digital, which, considering _most_ of my classmates and seniors in VLSI are doing, seems a more 'charted' territory
Nice analysis. In my view, we can further divide the field into 'needs the best performance' and 'works well enough'. I think for now analog designers have job security for the high performance area. However, I believe that we'll see the growth of automated and AI tools for 'good enough' analog. Especially in the open source field, as it's a lot cheaper to test and develop.
One of the criticisms of open source ASIC tools I've heard is that the older process sizes are not relevant, but in fact they are still very highly regarded by analog engineers for the reasons you outlined in your video.
Additional interesting reading can be found here: "Open-Source and Autonomous Temperature Sensor Generator Verified With 64 Instances in SkyWater 130 nm for Comprehensive Design Space Exploration"
There's actually quite a few other reasons why newer processes aren't well regarded by analog designers.
First, those transistors are often optimized for good digital performance without any regard to analog performance. This means they might have things like a poor gm curve. Or lots of equivalent output resistance. Or bad subthreshold leakage. Or poor matching characteristics. Or just plain weird analog behavior (like gate oxide tunneling). Or...
Second, those processes are rarely setup for good analog circuits. So, they will lack things like the ability to mix and match transistors with different threshold voltages. Or perhaps lack good resistor options, particularly high sheet resistance or low temperature coefficient resistors. And stuff like BJTs (yes, still common on analog designs, especially references) is virtually always an afterthought on those processes. It's usually some parasitic device with stuff like really bad betas (current gain from base to collector) and lots of undesirable parasitic devices and leakages. And zener or schottkey diodes are virtually nonexistent on those nodes. They will also rarely have things like extended drain devices for high voltage operation. It's also often really helpful to be able to put devices in different wells, both for isolation purposes, but also so you can float transistors above (and sometimes below) the substrate and that ability will be limited on a digital process or come with major drawbacks.
Third, alot of analog circuits go into power converters or RF. On both of those, it's rarely the active analog circuitry that dictates most of the die area. For power converters, it's usually the power switches, which will dominate the die area and don't benefit much from ultra small geometry stuff. For RF circuits, it's often stuff like inductors and transformers that dominate the chip. The actual analog circuitry is typically tiny, then you usually have some moderately sized mixed signal stuff to digitize the analog, then a big digital block to process it all (often on a different die - usually on a process optimized for digital). A good example would to pull up a die shot of the ESP8266 (a low end microcontroller with integrated WiFi). Zeptobars has a nice one. The digital is actually a pretty big chunk, but the single most prominent features are the big inductors and transformers for the RF (usually large octagonal or circular constructs).
Fourth, most of the actual analog processes get less attention and less money. So they are often old digital processes that have had analog stuff bolted on after the fact. Not only does that make them less than ideal for analog, but it all but guarantees it will be an old process. Which is usually fine because the analog rarely actually needs stuff like finer geometry (doesn't actually make the chip that much smaller), so why pay for more expensive masks and processing? And why incur the MUCH more difficult design rules that come with fine geometries?
How could you commented 3 month ago? this video is 1 hour old!
This comment is 3 months ago ?! how come ?!
Wow.. that's a bit creepy.. How does it feel to be a time traveler @matthewvenn ? Is the trip backwards through time (and presumably forward through time.. if you're reading this) as nausea inducing as I imagine it to be? Or is it like surfing a gentle wave for a time traveling bad@$$ such as yourself?
Analog circuit design using discrete devices is an art by itself. Placing an analog circuit onto an integrated circuit is art at the grandmaster level.
As someone who has *tried* to use the Berkley Analog Generator (BAG) one of the main problems is setting the system up. First it is a python library (built by grad students) with lot's of dependencies, these are not always readily available on the operating systems used to run EDA software (*cough* RHEL 7 *cough*). After that hurdle is cleared, BAG needs a set of base layouts that the software can array copy (for lack of better words) next to each other. These base layouts are necessary to build up larger transistors with the flexibility to meet the design needs. These layouts are difficult to make due to the complicated design rules for current-gen and leading edge processes. Also, the work to build the layouts cannot be shared between processes as they are, in general, quite different from each other.
All this to say, BAG has significant adoption costs associated with it.
Just to clarify, there's no metal in the center of the FinFET, that's silicon--an even worse conductor (semi-conductor in fact), which means even more resistance.
Been part of several chip and board bring-ups. Getting the analog wrong through bad or lacking simulation cost us a lot of time and money.
20 years ago somebody brought me on as a contractor to add digital control to an otherwise analog system. There was zero chance to make it work as they had not bothered to properly characterize their design and understand the component constraints needed for reliability and EMI/RFI compatibility & stability. 🤦🏻
Working at an analog company we only considered ESD protection the black magic. Most analog design was fairly automated after proper modeling.
Qualifying a new process was very in depth, and there were a few gotcha’s that popped up in the first designs with nearby devices or interconnects but those could be worked in design rules and coded into automated design tools. We would occasionally find a new problem in mature processes, but those learnings were quickly tested for in other processes and proactively prevented.
We weren’t at 7nm, or 32nm, but the analog circuits were always larger geometry then the digital portions even in analog heavy designs.
Put thermal accurate thermal models behind each component and a spice of minority model of the substrate connection it turns to less an art but more engineering.
Your presenting voice has gotten so much better dude, your work had been super thoughtful and now you're coming into perfection
1:00 It is no longer limited to "comes from real world" QPSK, QAM are extensively used in real digital world to transfer data more than 1 bit / cycle - >10 bits per cycle. Not to mention SDR radios etc.
All digital is analog - just depends on how deep you dive.. Thanks for what you do.
15:30 "I hope it won't happen. Because analog is the real world, and its designing is an art" Amazing conclusion of video emotion, kinda touchy
I worked as an FPGA digital designer with an office next to group that did mixed-signal chip design for imaging readout chips. This was a little different from some of the examples in the video because the imaging chips were and are much more analog than digital, and it was no secret that these mostly analog chip designers were the technical stars of the company. I had the privilege to be asked to sit in on one design review and was mightily impressed with the thought and skill that shown through in that presentation and discussion.
Thanks for being one of the best knowledge communicator out there dude.
This must be my favorite episode so far! I especially loved the ending. And the cake, too
I love your semiconductor videos, as I was a product engineer, we worked with both process engineers, design engineers, and test engineers.
It’s kinda interesting to think that even digital circuits are technically analog at their physical level. 😳
but then you realize that all circuits period are digital at the most fundamental level.
I had this thought too.
@@homelessrobotNo, they really aren't all digital. Up to 20% of each digital clock cycle of many architectures is spent in the linear range of the transistors (this is a large part of why they draw power, and why higher clock speeds draw geometrically more power) and not in saturation. Transmission lines, may they be data busses, networking cables, or power distribution wiring are all also inherently analog.
@@DrewNorthup what I was refering to is the planck constant. The energy of particles being in discrete increment, and time passing in discrete windows within which no ordering or subsequence of smaller moments can be discerned.
@@homelessrobotIgnoring spin won’t save you.
If Bob Pease were still with us, he say, "What's all this AI stuff?"
Great stuff as always. There's been a change of tone compared to your earlier work which was more academic. I suppose I prefer that, but I am glad you are having more fun with these! Please keep this stuff going.
Hi. I am engineer. Analogue design will be completely automated, if not already in some heavy projects, because the design reuse has reached its maturity early 2000, i.e. now there is only the need for advanced heuristics and optimization algorithms to achieve a middle range complex design of about 10000 transistors (we are not speaking of digital designs since these have pretty fixed rules(Lynn Conway anyone?). Unfortunately, the problem is that the new generation is not interested in analogue design, as being confident that digital systems will prevail or even they can outsource their analogue designs in the East. A bravo to Asianometry for this great presentation.
Indeed improving the reuse is where AI can help
I worked for years in the EDA SW team at HP/Agilent working with Test & Measurement teams. You are not kidding, analog design, especially very high frequency analog is freaking insanely complex & difficult...
Analog design is the Gordian knot of design disciplines. AI as Alexander Invictus however will be denied a single sword to slice through it. It will have to unravel the knot every time. If it can, let everyone beware.
@corbinauriti7807 Personally, I think I would beware in either situation regardless
My first analog circuit was a DC/DC converter with multiple inputs, single output to define the desired output voltage given the multiple inputs (it had power delivery limit, desired speed and temperature), it was designed to move a DC motor for a prototype car. Why did I made it analog? I dunno, I just loved the idea of making something different using something no one around me was able to even try to copy
Touching, funny, deep! Anyone with even a passing familiarity with circuit design can appreciate this! Thank you for brining this human element to a highly technical subject! Kudos and godspeed!
Love your videos, always nicely structured and explained well.
Greeting from southern Black Forrest (Freiburg, Germany)
As someone working in IC design, I congratulate you for the accuracy and clarity of the video! I'd just make a very minor correction: verification starts before, and is done in parallel with, the physical design. It only finishes after it, because it needs delay back annotation .
As newbie analog layout engineer (not circuit design) the main time-consuming part for me with a new pdk is studying the DRM and making sure all the rule checks/requirements from circuit designer are cleared after the design and of course all that PEX/EM/IR/ERC issues and what not are met. My job sometimes feels like laying bricks following specific rules after the main designer (circuit designer) has designed the stuff for the building. I think the layout part could get automated at some time in the future (in fact some things are already done using scripts that does not need human intervention).
Also, fun fact: Major companies usually outsource the manual jobs to 3rd world countries like India/Bangladesh. However, apple still maintains a very strict policy and their contractors have to be (I believe) physically present in their country/office. No wonder apple still rules the chip market with their efficient hardware-software combination, and their chips are so optimized whereas others are 'if it works it works'.
I only randomly view your videos, but it was cool to hear that your dad designed analog chips and that you were able to explore his realm with this video.
Yep. We had implemented everything in analog manner before this digital technology era. It was extremely hard if an implementation had to handle very high frequency signal. So, I feel that the analog tv systems were a miracle now a days. Anyway, thanks for your great shows.
4:30 - Like the old transistor radio schematic !
Those who aspires to be analog designer should have many first-hand experiences on discrete high frequency analog circuits design first in early age. You can experience many varied and weird parasitic behavior of analog circuits on such RF circuits. Familiarize yourself with many historic analog circuits design like radio, TV, classic amp, radio transmitter, op amp circuits. Study workings of the circuit carefully down to a resistor. Once you are accustomed and learned such strange and intricate nature of analog circuits, then study carefully circuit, layout, and floor plan of famous classic analog integrated circuits like LM741, NE555, LM324. etc. Compare SPICE simulation result and real circuit behavior. Know the difference of SPICE mode and real world integrated circuits. Try to understand why they are designed circuit and layout such way. You will get lots of insights from such classic designs.
Hey, they should learn vacuum tubes first ;)
@@stevebabiak6997 Actually it is good idea. Classic vacuum tube amplifier circuit has quite well-designed linear circuit. FET (not bipolar BJT) works like vacuum tube. So, you can learn and borrow many ideas for MOSFET circuits from early vacuum tube circuits. 5 tubes superheterodyne vacuum AM radio receiver is good example to study to learn about analog electronics. I myself leaned from classic vacuum amp and radio.
@@youcantata - correct, FET and vacuum tubes have some similarities. But there are cool devices too, like tuning eyes for example - that don’t have a direct analog in the semiconductor world.
Thank you for great episode. I can grab the picture and challenges of Analog chip design now.
It might sound weird, but capacitors are analog. And tiny little capacitors make up RAM. We try and clear them, and recharge them, within tolerances, but it's definitely not a binary value between those two charge values.
All devices are analog. It is only the interpretation that we put on the signals that is digital. But whether a circuit's signals are destined for analog or digital interpretation places very different constraints on its design.
Memory cell uses cross coupled devices tied into a positive feedback loop. To write a memory cell, you need a charge pump & memory driver(both are analog circuits).
As @danmenes3143 says, everything is essentially an analog device, simply because it exists in the real world. We just truncate that function to digital, for ease-of-use and manufacturing.
I wouldn't be surprised if we end up with pseudo-super-conductors, where it's not lossless, just really fast and efficient, after more research into the art, science, and craft of analog circuits and devices.
This is true for all electronics. "Digital" electronics just means they work close enough to an ideal digital circuit that we can assume it's digital. This is not always the case though, case-in-point for your example of DRAM, the Rowhammer attack, which repeatedly wrote to strategically chosen memory locations to influence other locations in memory.
"Every idiot can count to one"
Bob Wilder, the designer of the first mass produced OpAmp.
Bob Widlar, and that classic quote, got a mention in Jon's video about National Semiconductor a couple of weeks ago. :)
@@Gameboygenius - he was mentioned in this one too I think.
Nice coverage of the topic!
The core of the talk was about analog layout automation, yet there are many analog design automation methods (including ML based approaches) worth mentioning.
Your channel has some of the best videos ever covering semiconductors technologies/manufacturing. It's great to know that all falls back to your father being an analog designer!
Yay! I love analog circuit design, though strictly as a hobbyist. I have discovered areas where analog circuits still perform better than their digital equivalents, for example fast PID controllers. The closing sentence sums up well my feelings about it being art.
Didn't expect 'that' kind of numbers 1:30 in videos lol
10/10. Your style is art, too.
Woah, it never occured to me that physical designers for digital logic lost jobs, I thought most of them just switched to working the CADs instead of doing that by hand.
Analog chip design is very tough. This is a great video that can be touched on multiple times. Safety critical devices pretty much only take analog inputs and as we innovate some of the new boards are not as good as the old ones from the 80's !!! We moved backwards in analog chip design !!
You left out power supplies, which are a major block of analog on soc parts. Analog capacitance and magnetics are physically big, so they take up a lot of space as do the deep wells. Rf design such as wifi cellular and Bluetooth also have considerable analog preprocessors, but are part of most soc's. The topology of analog can switch from unbalanced to balanced in integration, this driving up complexity. In the end, I don't hold out much hope for ai, but a combination of ai with quantum computing that can model multi dimensional parasitics, now we are talking.
1) AI is trained on data from the Internet.
2) AI outputs data to the Internet.
3) Goto 1
... Exponentially growing AI-feedback ... you now see how this ends :(
A similar thing that has always baffled me is analog mechanical computers like the fire control system in Iowa battleships. Which as far as I know remained a lot more accurate in it's firing solutions than many digital types that followed. And since it mixed inputs continually it also gave you solutions real time. The only not real time part was the human operator needed to feed the wind speed or some of the other inputs into the system. But other than that it was solving an analog equation with the given inputs.. mechanically.
As a student of the binary age I've always thought that was really crazy that someone thought of doing things like that.
Yea, I watch Ryan over @BattlshipNewJersey as well. Digital is like quantization - it's discrete. Analog natively supports 1/2, 1/4, 1/8... 1/n.. 1/n*2... increments. Terms like a "smidge" are perfectly valid. Also, *all* of the inputs on the Mk 38 Range keeper along with the Mk 41 stable vertical are analog. Distance, bearing, temperature, humidity, on and on. Analog calculating analog is ideal in this situation. The inputs are the impetus for the outputs. The output changes continuously with the inputs. The firing solution is a natural result of the inputs. It was 40 years of development before the analog to digital encoders, and the digital computers were fast and reliable enough for the Navy to make the transition. Even then, the firing solution was discrete. Inputs became fixed so a firing solution could be calculated and made available when the trigger is pulled on the pineapple.
@@fredinit And 5hey were using analog firing computers in the 19th century in the early battleships and cruisers.
Having 40+ years of electronics repair and design under the belt, I found this video fascinating and enlightening.
Thank you. 😊
This video is exactly what I needed after a talk about this with my mixed-signal circuit design professor
I started my analog design career with a CK722 transistor which I promptly burned out. I did a lot of analog designs using analog chips like the 741 op amp. I got into digital electronics before TTL came out. Went on to use the first microprocessors but they were too slow to do much of anything. I invented a digital circuit design plan that I called 'random logic' which implemented software using hardware. My random logic was much faster than any CPU could do.
Just like any truly creative art. AI is a tool to be used. It's really handy. It (for now) cannot replace the nuances of human. AI can conjure unexpected things that might work with a little tweaking from someone with experience. Saves a lot of trial and error. Doesn't replace the human. Human can do more in less time. Positive feedback loop.
Remember the bad video transitions of the 80's and early 90's? Yeah. That will happen and then people will hate it, stop buying it and it will go away. Then it will make a come back for a while and then be regarded as lazy again. The real art is timeless.
Love your channel big bro.
I used to be more of a software type of guy, but you kinda nudged me into messing around with hardware alot lol.
I might just get an FPGA and see if I can make my own stuff in the near future lol
Based and veriLog pilled
Same, I got inspired reading koyotekola6916's comments on the FPGA video :)
Today, they employ various building blocks to create larger groups or modules. These building blocks include oscillators, switches, amplifiers, and so on, which are assembled into functional units. However, the approach taken depends on the specific RF (Radio Frequency) components they are working with. When dealing with GHz (Gigahertz) parts, the design process often starts from scratch, with components being integrated into a complete system. In contrast, for low-frequency parts, the approach typically involves assembling pre-existing modules and grouping them together to form the final system.
Didn’t Analog Devices once use the catch phrase “we live in an analog world “?
If you're ever in the NYC area you should try the black forest cake from Rudy's in Ridgewood.
This used to be a lot more manageable (not necessarily easier) when I started electrical engineering back in 1980...
Quite a good video. It reminds me of Keith Elliott Barr book "Asics design in the silicon sandbox".Best Regards from France.
Something not mentioned in the circuit layout part is the inability for algorithms to route circuits in a determinable amount of time. Most algos used in programs have a best/worst/average case (BigO notation) that is polynomial which mean time to complete is an exponent of the items to iterate over. Some algos are deterministic but some, like the travelling salesman problem, are nondeterministic and thus cannot predictably be solved in the most efficient way possible by a machine.
That's exactly the point if the parallel with the game of go.
If you can create a Go champion with AI, you can do the same with analog circuit design. It's "just" that nobody's thrown the money and effort into it (yet).
Anyone here remember the OpAmp 555 timer?
Yup. The 555 was the heart of many projects back in the day. That and the 221(?) dual monostable.
Never remember hearing about that, no. I know about opamps and the 555 timer but not the opamp 555 timer. 🤔
I remember well as it is often used in our Logic Labs during those study days.
@@Gameboygenius Well, 555 is not 741. But, there are op amps inside 555.
@@JohnnieWalkerGreen Kind of? They're designed to be comparators and probably wouldn't make very good opamps because they're too sensitive.
"delicious black forest cake"... "oh wait, where was I?"... I love this channel.
nice :) I really struggle with the analogue parts of my projects :) I notice quite a movement to re-create old analogue chips like sound chips with FPGA emulation based on digital filters and some very creative bitstream shenanigans. That is open to AI approach too I think ? I recall some mobile robot gait generation approaches based on AI gen of bitstreams for FPGA ... with class D amps all sorts of oddness become possible on the analogue generation side.
I would like to know all about analogue sound chip FPGA bitstream shenanigans, if that's actually what they are and not just clever digital programming.
You might be aware of Adrian Thompson's classic 1997 paper "An Evolved Circuit, Intrinsic in Silicon,
Entwined with Physics". He applied a genetic evolution approach directly to the bitstream to evolve a tone detector that could discriminate a 1 kHz and 10 kHz tone on a FPGA with no external timing reference. It got there eventually. But the odd thing was that the circuit relied on cells that were disconnected to function. And when he tried moving the bitstream to a different of the same model, it no longer work. The genetic algorithm had evolved to live on one specific piece of hardware.
You need to interview guys like Doug Curtis, Dave Smith, Tom Oberheim, Dave Rossum... if you want to know about analogue IC design. Sadly Dave Smith passed last year but the others are still around and putting out products based on analogue circuitry.
10:20 "Shrinking metal pitch from 80 nanometers to 48 nm raises the line resistance by 6 times" 80/48 = 5/3 = 1.66 (5/3)^4 = 6.4 if this exponent continues to 10 nm -> (80/10)^4 => 4096 times larger... crazy
As always, great video. But you forgot to mention the most exiting one: parasitic inductance.
It really can do some fun stuff together with parasitic capacitance 😅
And for all the people that are impressed by the diagrams shown in this video: the white space… that’s were all the parasitics are hiding!
Study in Electronics engineering almost 30 years ago but didn't work in this industry after that. This video reminds me my old day. The day I can understand and have to calculate each bloc to double check things. Talk about Spice ... Man I really miss it. Another software that comes up in my mind is MATLAB.
Analogue electronics is where its at for me personally, I will continue to try and learn it till I die...cheers.
As a hobbyist and looking into pcb design I was hoping AI would help with analog design. Looks like I’m going to learn the hard way which is entirely much better any way because it forces you to understand the intricacies of the design and allows you to troubleshoot when needed. Cheers
Astrus? Whoever naming that company is a genius!
Imagine people introducing them ,may be on TV or stage : "This's Astrus."
I like how you kept the cake for as much time as possible on screen
The idea of an analog life form remaining the only means to design an analog chip would be poetic.
Famous last words... xD, but whether they are mine or yours I do not know.
I personally believe that AI is quite suited for design processes with a lot of variables, although humans will always guide the process to some extent, but on more and more abstract levels moving forward. So with less input garnering more control.
I think the key is connecting an LLM to a generative AI and an automated prover. The generative AI can adversarially learn to move across the state-space given by the proving process (it mimics validity). If that proving process includes certain effects, the generative AI will be capable of handling those too. This generative AI can be initially setup to be partly driven by some huge random input value, which then can be replaced by an LLM setup to influence the design outcomes. The LLM learns to output values that cause the circuit design to mimic those it learned elsewhere. A good start would probably be all the textbooks about circuit design. This should enable high-level parsing of abstract input (and iterative guidance of the design process).
excellent video about the difficulty of analog circuits. I'm hopeful someone can train a model to make this easier. I have to think TSMC is working on this problem internally.
As someone struggling and constantly befuddled by simple circuit designs... A lot things clicked watching this. 10/10 quality content and knowledge on this channel!
Ah,yes, design a Butterworth filter for the following frequency range, with a Q TBD.
Excellent overview Jon. But don't let my review go to your head, please.
ASML / TSMC probably need to replace the copper with silver/ platinum/ gold or alloys thereof, for the sub 3nm 1nm & so on 0.6nm nodes since copper parasitic resistance in shrinking pitch sizes increases so rapidly as a function of the cross section decreasing.
This channel is a true gem, thank you for these awesome videos!
Can you elaborate on the issue with the proprietary PDKs? Looking at the GitHub for both align and magical they appear to be BSD licensed, so there shouldn’t be any issues there, do the PDKs have some kind of anti-OSS licensing associated with them?
Analog circuit design is an art, if good engineers designs it with understanding, of what the circuit has to do and later tests and layouts it.
If neural networks do it, you likely get a mess, that somehow works, but with no understanding in how and why.
But AI by itself would just put out garbage. Analog Design encompasses a wide range of fields: PLL to AFE to ADC to switching regulators to RF design. Too broad for AI.
This is probably why the latest A17pro is not quite what we expected
I'm a senior engineer. I also see analog design as an art that needs time to master, which is being lost in a world were something that has more than 5 years is too old. Thank you for your kind words.
I would love if you made a video on the Semiconductor supply industry. like who makes the machines that make the parts for the semiconductor industry. i work for a CNC dealership. Matsuura is a great company and they have this really cool metal 3D printer that also mills at the same time
I am lucky to be an artist of this art. And I hope it won't get automated anytime soon.
No need for hope, you can trust
The numbers... Yes... I recognise some of those numbers...
Thanks for analog art form review designing with pencil and paper
I still design with vacuum tubes and transistors.
Your videos never fail to reignite my passion for engineering, awesome stuff!
@asianometry Good to know that me as real world useless magister of philosophy came to the very same conclusion as a technical expert like you = "Life is analog".
Exactly one year ago as a notorious coffee loiterer and registered inventory I ve experienced transcendental rational blackout and stroke of enlightenment of my own ingenuity and I opened my a own coffee house (bar) called Barista (what a original, doesn't it) with main claim: Barista - "The analog type of coffee house" and often added with subtitle: Real life is analog...
So in conclusion.
Despite:
a) I am totally out of target group,
b) knowing about semiconductor chip design less then amount of milk in ristretto, and...
c) main goal of this video is to create ideal circumstances for tech savvy monsters to cause them at least week long chain of wet dreams...
Despite all af this, I really like this video :-)
My first thought in this video, was correct selection from about millions of existing semiconductors in our world in designing analogue circuits boards. AI could help a lot in selecting these in function and low costs. This could speed up designing processes. // Would you do the same video in analogue design of elemental digital circuits components? => Design rules of transistors, gates, flip-flops, synapsis, neuronal circuits, ... ? Thank you for this video.
It's a bit poetic since, accdg to a Veritasium video a few months ago, analog computing may make a come back as these maybe better for AI.
I love this line "Digital world is all about numbers, a concept that human make up"
analog designing is to ai driving a car versus to a human, ai can only work on syntax but human can work on both syntax and semantics. in less nerd terms, ai knows the best route to take but humans will take a route because he needs to.