People are competing on how to apply analytics, data mining and machine learning in huge markets and applications, and semiconductor design and manufacturing is undoubtedly the most promising area. Let's take a look at the related content with the embedded Xiaobian.
The key to machine learning/deep learning/artificial intelligence (ML/DL/AI) is to understand how devices react to real events and stimuli, and how to optimize future devices. This requires screening more and more data, automating to identify complex patterns, anomalies, and finding the right place.
MikeGianfagna, vice president of marketing at eSilicon, said, "The data we collect is used to develop our own methods. For current storage, we look at design, memory, and modeling different memory configurations. You can run models and real applications on it. For comparison, we use a parametric generic memory model and map it to real memory. We can also look at the manpower and timeline from the previous design, and do the same for computing resources and EDA licenses. If you only have 12,000 CPUs but 24,000, then you need to use a cloud-based solution. But that won't happen very quickly, you need to plan for it, a lot of it is around memory."
This is just the beginning of a series of activities that shorten the design cycle and reduce potential problems based on experience.
Michael Sanie, vice president of marketing for Synopsys Verification Group, said, "The industry is learning how to use artificial intelligence and machine learning to build and debug systems. There are a lot of stacks that can be used for modeling and simulation of AI algorithms. The goal here is to artificially Smart for tools and other places."
To a large extent, this is an understanding of system-level design and complexity, even if these tools require the help of other tools.
StevenLewis, director of marketing at Cadence, said, "Now, with the EDA ecosystem working on machine learning, we are trying to figure out how it can help us solve larger verification problems, and the whole problem is back." This is not early. 10 transistors'. I agree with this. To make a memory with 1 billion transistors, you need to know how to layout, how to route at the physical level. How to perform the best layout on these circuits, how to place these components? It has always been part of the machine learning task. However, we don't call it machine learning, but it is really part of the algorithm we are trying to find the best way to do it.
If you can handle a specific topology and start to understand how 7nm transistors behave, then you know most of the flow, so you can better determine when to start placing them, when to start layout, and when to start analyzing. â€
Some time ago, chipmakers began to promote six-sigmadesigns, but once they were so complex that the simulations required to achieve the required six sigma quality began to take a lot of time, they would not discuss the solution. But with car manufacturers' demand for flawless electronics, and the discussion about Six Sigma, the only way to achieve this in a reasonable amount of time is to use machine learning.
Lewis said, "By machine learning, we can program the behavior of the transistor, so from a statistical point of view, I don't have to do 10 million simulations. I can use algorithms and machine learning techniques to maintain certain accuracy and determine the need to run. The minimum number of simulations. If I can program more algorithms, if I can program more of my work, then I can only collect this data while still guaranteeing the quality of the work. This is what machine learning can help us. local."
This has happened to some extent in the design field, but it will be more significant in the future.
TyGaribay, chief technology officer at ArterisIP, said, "This technology allows you to implement things that were not possible in the past. You can use this technique to determine the location of the vias. If you can't prove something, you can use the exclusion method. Use the tradition. In the form of functional safety, you can put down two identical things, just like in car brake systems, engine controls, airbag control systems. But when we drive through wires, we usually create a server in your car. Integrate functionality into the system and communicate via wires. Currently the 777 jet can do this, but the cost/function varies widely. In order to apply it to the car, we need to develop new technologies and find applications to apply them. Method, replacing technology that has been proven in the past but is too expensive and very slow."
massive dataThis will require filtering a large amount of data.
AnushMohandass, vice president of marketing and business development at NetSpeedSystems, said, "In Silicon Valley, almost everyone is good at getting some form of data. But the really powerful people are those who have gone through a complete chain, and they understand the data chain from understanding to action. We have put this back into our design environment and IP. Machine learning is one of them."
Mohandass said that the key to using machine learning is to mine first-hand data. “What is the training data you use? How big is it? Are there deviations between them? Are you biased towards some form of design or other design? We spent a lot of time to speed up our training without any bias. The two aspects are the amount of data you get, how do you understand it deeply? The third is our machine learning environment, what is the driving force? How to drive customers?
When generating data, especially if you type into the machine learning engine, the amount of data will explode for you. You don't see hundreds and thousands of rows, but millions of rows, so it's meaningless to suddenly see a chart of one million things, so you try to get them together and observe their trends. â€
In the case of data mining, a very practical use of machine learning is to search for anomalies.
UltraSoC CEO RupertBaines said, “We have done a lot of local work, so the results are very valuable and the content is very good. In most cases, this means that an engineer is looking at it, reading and drawing it. Sometime, she would say, 'It looks weird' and will run some scripts to try to find out why it looks weird."
By helping them find problems and getting their attention to a sudden change in behavior patterns, it can help speed up the engineer's work.
Baines said, "This may be something like a constant pixel in a camera app. You say, 'It looks weird', you expect the pixel to change. This could mean that the camera is malfunctioning, stuck. It could be a safe An application that never accesses a specific process of secure memory, and it will try to do so, the anomaly detector will say 'this is wrong, turn the lights on and issue an alert. At least on the other hand, an alert will be issued. This It's like a burglar alarm. If someone is pacing outside your home at night, you have a light sensor that can detect their movement, when it turns on the light and gives an alarm - even if they don't really try to enter, even if you The locks are still intact, they can't get into the house, but you know someone is trying to do something. It's an extra layer of security on top of TrustZone, OmniSecure or something like that."
Machine learning and simulationAnother area in which machine learning is being used is to determine if the simulation is repeatable if the simulation is run next month or if the simulation is compared to other simulators. This also applies to manufacturing tests where the periods of the results are related and the signals are related.
MarkOlen, product marketing manager at Mentor, Siemens, said, "The challenge of machine learning is that you give up the ability of humans to control things. This is the key. If I use machine learning technology with portable incentives on my 1000 CPUs. Simulations, which produce a set of results based on how the design responds, because it learns from the design's response.
"But assuming that after running for a few hours, we found two bugs in the design, we sent the design to the designer, they repaired the Ethernet module or repaired the structure arbitration scheme or something else, and then run the simulation again. If you use machine learning again Simulation, you will get different results. You won't get the exact same inter-cycle correlation, because the design runs differently, probably because of the repair of it. However, this creates a lot of uncertainty for engineers because they Indicates that they want to run the exact same stimulus under exactly the same conditions, but the design has actually changed.
“Because of this, we have switched between some of our technologies that can turn off some of the machine learning features so that we can run some kind of simulation mode that we did before. At the same time, there are some advanced customers who are The idea is satisfied and is using all the features."
As the design team runs the simulation repeatedly, it will enter the data mining of off-chip indicators. Here, what is useful inside Mentor and its customers is an open source software ecosystem called Jenkins. Olen said, "This is a very hot topic, everyone likes it because it's free. But it's not completely free, because even if it's an open source ecosystem, you have to invest as a user to actually do it. Integration. We invested a lot of money to integrate our system into the Jenkins environment."
One of the main advantages of Jenkins is that it can be like a trigger. He said, "You can have a timed trigger to show that when everyone goes home on Friday night, there are 10,000 desktop computers idle, let us use them. So, at 9 pm on Friday, you will start to return free of charge whether you need it or not. Then, once the results of these regression runs are complete, we can combine all the results and send them back to Jenkins so that it can appear in an email from a vice president at the engineer's desk on Monday morning, 'The good news is, We did an 800-hour simulation with no failure. 'Another trigger event is not based on time, but on conditions. For example, if there is a certain level of code change in the design, or every time you modify a file, you can start a regression run. You can speak automatically at the end of 8pm. If more than three files are modified during the day, then you will start running back at night."
This will run back after the run and the peak of the data. This is not transaction data. This is not about bus acknowledgment, fetching, or what actually happens on the chip. Instead, this is all the results of the simulation, such as detected errors, coverage, and periodic runs. But one of the problems is that it is necessary to figure out how to combine multiple types of data.
1000 puffs disposable vape pen are so convenient, portable, and small volume, you just need to take them
out of your pocket and take a puff, feel the cloud of smoke, and the fragrance of fruit surrounding you. It's so great.
We are China's leading manufacturer and supplier of disposable vape bars, 1000 puff vape pen,1000 puffs disposable vape,1000 puffs vape kit
vape pen 1000 puffs,vape bars 1000 puffs, and e-cigarette kit, and we specialize in Disposable Vapes, e-cigarette vape pens, e-cigarette kits, etc.
1000 puff vape pen,1000 puffs disposable vape,1000 puffs vape kit,vape pen 1000 puffs,vape bars 1000 puffs
Ningbo Autrends International Trade Co.,Ltd. , https://www.mosvapor.com