I want to talk about my own growth process, mainly how to improve my technical ability. Let me briefly introduce it. I have been working on audio dac/adc, which is sigma delta dac/adc. In audio, dac is far more important than adc, so the focus is still on dac. At present, the level of dac is more than 100 db dynamic range, the key is that the power consumption is very low, the area is very small, based on the company's original 70 db, not only the performance has achieved more than 100 db, but also The consumption area has been greatly reduced, and the performance power consumption area has exceeded the well-known foreign IP suppliers. You buy something like mp3. Is it important to play important or recording? So dac is more important than adc. I am not a big cow, but I have a little bit of sentiment and I have to share it. Wrote well, I laughed. When I was at school, I always wanted to be a programmer and I was trying to figure out programming (it is very important for an analog circuit design engineer, which will be mentioned later). After graduation, I became an analog circuit designer. Actually my major is not a computer or a microelectronics. When I first entered the company, I only knew a little bit of the bandgap reference design. In fact, I didn't even understand the ac/tran simulation. I don't know that the circuit must undergo tran simulation. Miller compensation doesn't know what it means. Looking back now, I laughed at the dead. When I was just assigned to the project team, the supervisor was busy with the project. Just when the project encountered a bottleneck, he didn’t like to bring people. I felt that I couldn’t use it anymore. I just gave it a topic: study classd, it’s quite hot now. Research is good, you are an expert, no one has studied it. So I started a sheep-like career for more than half a year, and most of the foundations were established at this time. What are I doing during these hours? Look at Razavi, Gray's book, found that there is no relationship with classd, looking for paper on the Internet, read nearly 100 articles, or no clue, after all, there is too much garbage in the paper, find out the really good paper is not easy. I finally found a classd paper from TI and spoke very well, so I tried to make the circuit and started the simulation. Later I will talk about some of the ways I feel really useful for learning analog circuits. It’s true that the opportunity is still very important. The important thing is that you are good at grasping the opportunity. Life is like this, what time period to do things, if you can't grasp it, there will be no such good time in the future, and cherishing it is very important now. I have been working for more than 3 years so far. The IP provider is ChipIdea, which is now acquired by mips. Classd was still quite a fire a few years ago, and now it has become a cabbage. Many people have misunderstood that the performance of classD is not as good as ClassAB, and it is not comparable to ClassA. It is actually wrong. ClassD can still be very good. If you design well, the 96dB dynamic range, the performance above -80dB is very easy to get, and ClassAB is not easy to do harmonic distortion. ADI has an AD1991, the performance is very high, far higher than the average ClassAB, but I think its architecture is general, there is a better architecture can achieve the same performance, but the power consumption is lower, it is very important that the development cycle can be more short. The key is the architecture. In fact, all circuits are like this, the key lies in the architecture. It is very important to have 4 points in the circuit. 1, performance; 2, power consumption; 3. Area; 4, the development cycle The key to determining the tradeoff between these four ranges is the architecture. 1. A necessary foundation for simulation, the so-called foundation! Common tools for analyzing analog circuits: signal and system and circuit theory, these two undergraduate textbooks are precisely the basis for simulation 2. Deep understanding of basic analog circuit modules, such as current mirror, 2 stage miller OTA, folded-cascode OTA, etc. It’s not that I read Allen’s book, I understand the formula, but I can figure out the aspects that I have never mentioned in the book. Even the things in the book can have their own understanding and even correction. This process is not only a process that requires the theory to be strictly reversed, but also a process that needs to cultivate intuition; it is a process of reading books and learning others, and it is a process of independent thinking and forming its own design method. After all, the real innovation of making products, textbooks are far from enough. 3. On the basis of 2, form an intuition and experience value for the analog circuit design, which is the most useful in the actual design, such as how much a current mirror Vdsat takes, how much the area is taken, and the advantages and disadvantages of various circuits A deeper understanding, knowing that the required performance can be achieved with the simplest and most reliable structure (this is very important). 4. Accumulate engineering experience on the basis of streaming, knowing what other simulations are unpredictable in the actual process. This is often not a circuit problem, but may be a match between modules. 5. Further understanding of the circuit in the flow of the problem, and on this basis can be innovative. In the actual found a lot of simulation, the paper may have seen a lot, the book has seen a lot, but take their circuit to do a DC, found that many current mirror Vdsat only tens of mV, a differential pair working state is very Reasonable, such as input differential pair vdsat, active load Vdsat is not reasonable, such as the output level of the library L. There are still many people who use the artillery to make mosquitoes. It is a very common requirement, and it is a very complicated structure, regardless of the area. When I started looking at classd, I couldn’t understand it and couldn’t understand the principle. Square wave can represent a sine wave with very low harmonic distortion? Starting with an open-loop structure, that is, using the most ideal PWM generator, it is also difficult to generate a waveform with very low harmonic distortion. What about the non-ideality of analog circuits? At the time, I also saw the literature saying that it was time to do. I think how is this possible? Doesn't that create nonlinearity? But if there is no dead time, the power consumption is very large, and the EMI problem is difficult to solve. It’s too contradictory, there is simply no way to compromise. As we all know, the simulation speed of the switch circuit is very slow. It can simulate a cycle in half an hour, and the accuracy is not very high. Therefore, if I have any idea, I need to spend at least half an hour to verify. In fact, the time used is much longer than half an hour, because hspice often encounters convergence problems, and often does not converge when the simulation is halfway! Exhausted! It took a long time. Later, it was found that the convergence of spectre is better, and spectre has been used in the future. However, there are often problems with convergence. In order to solve this problem, I found a book dedicated to spectre simulation (written by the cow who developed spectreRF), how to calculate spectre, how to help the simulator to converge when doing circuit, and consider those when modeling. Convergence factor. This is the style of my design. If I don't understand the problem, I always go to a professional book or literature. I will study it seriously, even if I have to spend a lot of time to do it, the important thing is to understand! After that, I rarely encountered convergence problems, and even if I encountered it, I could solve it quickly. The contradiction of classD was not solved at the time. In the end, I thought that my idea was wrong. At this time, I still think about the problem from the time domain. However, when I studied the literature, I found the direction: 1. Solve problems from the frequency domain; 2. The focus of ClassD is noise shaping; So I started learning signals and systems. At that time, I felt that many people felt that the signal was important in the system, but there were not many people who really took the time to learn, because there were not many people who really solved the problem with the concept of signal. The habit of reading books is to read through it. It usually takes 3 days to 1 week, and then I will start to study the relevant chapters in combination with my actual work. But the signal and system are too theoretical, and I still don't know how to apply it to the actual situation. I think what ClassD can say is an analog filter, and I have studied the book of filters. In the meantime, I need to use the Laplacian to derive the transfer function. In the process of derivation, the concept of signals and systems is slowly established. But for the noise shaping of ClassD, I still don't understand. The focus of ClassD is noise shaping, which I learned from the sigma delta modulator. The sigma delta is a fairly mature theory of 7 C4 D, and few people are studying it now. I think, why not first understand sigma delta and then do ClassD? So I am looking for information in this regard. I happen to be a colleague of mine who likes to collect books on various analog circuits. He recommended me to Delta-Sigma Data Converters - Theory, Design and Simulation written by ADI. I now call this book sigma. Delta bible, most of my sigma delta knowledge is derived from this book. There is also a very important book: "Understanding Delta Sigma Data Converters" _ this book is also written by ADI's cattle. If someone asks me, what is the best way to do analog circuits? I would say that starting from sigma delta adc/dac is best because it tells you what is the signal, what is the transfer function, what is modeling, what is noise, etc. The core content of the analog circuit. If you want to learn what books to read sigma delta, I recommend these two books. The reason why I can make something in this field is that the two books are the biggest! But it's not that I will finish everything after I finish these two books. It’s still too far away, because it’s not easy to absorb knowledge. Moreover, to be honest, the circuit is not a textbook, and the book is correct. The reality is that we often don't know which ones are right and which ones are wrong. The paper is all over the world, but the wrong or ambiguous things are also in the world. When we encounter problems that have not been reported, we do not even believe that our derivation is correct. Many people often wonder after completing the circuit. Can the circuit they do work really work? Can performance meet the requirements? Is there a bug? Is there anything that has not been considered? The biggest difference between doing analog circuits and doing digital circuits or software is not the difficulty gap between the two, and the simulation faces too many uncertain problems, the software can run, the FPGA can run logic, but the analog circuit does not, only the ratio The snail is still slow in simulation software, and there is no way to simulate the real situation. It’s hard to have 100% confidence before returning to the film that you are right! The confidence to do analog circuits actually comes from how you can judge which simulation results are credible and which simulation results are untrustworthy. The real improvement comes from the debug of existing chips. At that time, the company just had a chip that was debugging, and there were a lot of problems. One of the problems was the problem of sigma delta DAC. I am in a meeting and other situations, I know the phenomenon of this problem, that is, the output actually has a triangle wave! This serious problem has not been solved for a long time. At that time, I got the design document and modeled the structure inside, but of course the learning phase. Speaking of matlab modeling, I have to focus on it, matlab is a very important tool. The advantage of doing sigma delta is that you must master this important tool, otherwise it is difficult to improve. As I said before, I originally wanted to be a programmer and wrote some code. At this time, my experience in writing code immediately helped me a lot. I quickly entered the state and felt very much about matlab, so I used it. Matlab began modeling the sigma delta system. Here I am going to introduce a book, that is, "Digital Signal Processing Practice Method". I spent a golden week reading this book. I don’t do anything else except reading a book and sleeping in a golden week. It ruined me, but this is a golden week that I don't regret the most, because I learned the key turning points of the signal and system. This book is very simple and straightforward. Many of the things I used to see signals and systems that I didn't understand are understood in this book. And the combination with matlab makes this book especially suitable for sigma delta modeling guide. This book talks about digital filters, not so much digital filters, but discrete-time filters. Isn't ADC/DAC a discrete-time filter? Moreover, I have also studied the algorithm of the digital filter, which laid the foundation for the design of the entire codec. This field is rather strange. People who do analog circuits think this is the field of digital circuits. People who do digital circuits are busy doing implementation. Without time to study, it becomes a vacuum zone. Only I am not busy with the project, I want to learn what to learn, I am interested, just happening to learn digital signal processing. So I studied it. With matlab as a practical tool, learning is still very fast. At the time, I wrote the code and optimized the algorithm. I have been reading the book "Delta-Sigma Data Converters - Theory, Design and Simulation" until one day, I saw the description of this phenomenon in the book, and the law and test results are similar, so I used this phenomenon Matlab has been modeled, and the modeling results show that the triangular wave can be modeled! And other phenomena are similar! So, I gave the modeling structure to the leadership demonstration. I said that the problem should be this, but I don't know where the circuit is going wrong. Give me time. I believe I can find the problem! What matters is not how powerful you are, but how much you invest. As far as your own experience is concerned, the most rewarding is always the time of maximum investment, and the amount of investment and the amount obtained are quite different. If you invest 80%, the income may be 20%, but if you add 20% of the investment, you may get another 80%. The so-called hunter is half the ninety. This is the truth. In fact, the phenomenon of the problem is different from what is said in the book, and there are even contradictions. In the modeling, I studied the possibility of the problem, guessed the possible line of the problem, and simulated the circuit. In order to do the simulation, I also studied the VerilogA language, carried out simple modeling, and studied the dft algorithm, and studied what methods can be used to improve the simulation accuracy and reduce the simulation time. The simulation of sigma delta is very slow, and it is often one day to be able to simulate a case. During the simulation, I was almost desperate, because the simulation results showed that the design was no problem and no signs of any problems were found. I constantly change the conditions to simulate, and I am constantly modeling. In the modeling, I found that this problem is unlikely to occur before a certain circuit, and this problem cannot occur after a certain circuit! Both reasoning and modeling illustrate this problem, but with a circuit simulation, I found that there is no such problem! Unable to reproduce. The mood at the time was: despair! The problem seems to be close at hand and far away, seeing hope, and there seems to be no hope! In this way, I simulated it for more than two months and almost gave up. In the constant simulation, I finally simulated this problem, but the performance and test results are inconsistent. At that time, my judgment was that the problem was not found out. I just found another problem that was not so serious. The debug process has entered a stage of despair. At that time, there were new projects to be started. There was just this module. When the module was sub-module, everyone was not willing to choose this module, but I was very willing. My idea is to find the problem before I start to make the circuit. In fact, it is very risky, the project has progress requirements, but there is no way to control the progress of the debug, it is easy to make the project uncontrollable. But the problem is not to find out how to ensure that there is no problem in the next version? Does the problem disappear from time to time? I decided to keep investing in the debugger and stick to the last moment before I started the design of this module. Just as the leader also attaches importance to this issue, he can continue to persist. When I saw the waveform someday, I saw the problem! There is something wrong with the waveform. I abstract this problem. From a signal and system perspective, it should match the test results. I built a new model, modeling simulation shows that the performance of the test results is consistent, the circuit has been adjusted, in the spice simulation, found that the problem is really gone! What excites me the most is: I have achieved the consistency of theory/modeling/testing/simulation! Let me be 100% sure, I found the root cause of the problem, the only worry is that I don't know if there are any undiscovered issues. After all, analog circuits and digital circuits are different. Digital circuit pairs are right, wrong is wrong, analog circuits are fuzzy, and simulations often have blind spots. Fortunately, there is no problem after the circuit is modified. To my surprise, the problem actually appears to be consistent with the early inferences. It is impossible to have this problem in the path before that point, and it is impossible to see this problem in the later path. The problem is at that point. This problem that has been plaguing people for so long is actually a small mistake. To fix this mistake, all I have done is to modify a line. But a lot of gains, far more than a line! The most important thing about sigma delta is noise. During that time, I got a lot of data, used the data, and modified the simulation method. Finally, the simulation and the test were consistent, and the confidence in noise calculation increased a lot. There is a saying: lowpowermeans understand more about noise, this sentence is very correct, know the noise, you know how to simplify the circuit, reduce the power consumption of non-critical circuits, and properly plan the power consumption of critical circuits. There are also many innovations that can be made to the circuit, including structural innovation. After the debug is completed, I started to work on the DAC in the project. I immediately made a comprehensive modification to the structure, reducing unnecessary noise sources, and planning noise and power consumption. At that time, the time was already very tight, but the speed of completing this circuit was very fast. Finally, I caught up with the progress of the project and tapeout went out. However, the project also died for other reasons. For the first time, the average person is not very innovative, and they all want to be stable. But my idea at the time was that innovation is the most secure and has enough confidence in itself. Looking back at the situation at the time, I felt very surprised. It was the first time I did the circuit, but I felt like I was an expert. I had almost no experience in filming, but I vowed that my circuit had no problem and I could reach it. Very high level. In fact, apart from myself, other people don't believe much, especially the old employees who have several years of work experience. But don't blame them. If I am in their place, my thoughts may be the same as them. I did not expect the post to actually add the essence. I am more stubborn in doing things. I want to study a lot of things. I don't like the state of understanding and understanding, so I spend a lot of time to learn. This is a slow-looking learning method, but it has taught me a lot of knowledge that others have not learned. After the above mentioned project is completed, I went back and did a classd project. It took a week to define the top structure, calculated the signal transfer function, the noise transfer function, and built the veriloga model. Various non-idealities are studied in the model, and a spread spectrum algorithm is also added. And simulation verification (both modeled). In the system, it is guaranteed that the non-ideality of the circuit has little effect on the performance of the circuit. In this way, the difficulty of circuit design is transferred to the system function instead of the analog circuit design. The same is true for transferring circuit difficulties to digital circuits. After I finished, I found that classd is a very simple thing, and certainly can achieve relatively low power consumption, and finally, so that I have no interest in even doing classd circuits, because in my eyes, this is a fully demonstrated The topic, there is no need to vote again to verify. Later, this circuit was handed over to other colleagues because I had a new project. In the end, this project was made and the performance was very good. In contrast, the performance of TI's classd can only be said to be very general. To sum up, the classd technique is exactly the same as I originally thought, using noise shaping technology, but it is simpler than the sigma delta ADC/DAC, because there is no real digital-to-analog conversion part, the circuit implementation is very simple. Original post address: http://bbs.eetop.cn/thread-141642-1-1.html Finished address: http://bbs.eetop.cn/thread-313574-1-1.html Watch EETOP, enter one hundred chest backstage view Recommended Reading: Analog digital product development process A brief history of SPICE Dry goods sharing: measuring your own ECG (from theory to circuit) Good analog IC engineers should have the literacy Classic in the field of analog IC design Do you need post-simulation simulation? I see the pole zero Six classic analog IC books wonderful comments and summary 100 Bibles designed in simulation Suggestions for getting started with analog circuits Analog IC film experience sharing Simulated IC annual salary of hundreds of thousands of brothers' model learning experience What should I do to become a simulated ic designer during my undergraduate degree? Nine-level advancement of analog circuit design Where is AnalogIC difficult, structure? parameter? territory? system? Analog IC Design Part 2: Transfer Function, Zero-Pole Formation and Time Domain Response I still don't want to continue to learn analog design - a question of inter-disciplinary master How to learn analog IC design Simulating the big cow to talk about the price and development direction of the simulation engineer Analog circuit should know 200 questions Several practices and advantages and disadvantages of post-simulation simulation Very interesting explanation about analog land and digital ground Dry goods! In-depth explanation of the classic lecture notes on RF analog circuit design The Road of the Modem--The Analog Circuit Learning Feelings Is analog IC design really hard to learn? Analog IC forward design process summary Simulation learning sentiment and summary Analog filter design handout Lazard Analog CMOS Integrated Circuit Design Lecture Part3 Lazard Analog CMOS Integrated Circuit Design Lecture Part 2 Simulation learning sentiment and summary Longkou Libo Insulating Material Co.,Ltd. , https://www.liboinsulation.com