Smaller analog chip companies have once again opened their doors to custom analog chip projects, which are too expensive for most chip manufacturers.
To meet the growing demand for specialized custom analog chip designs in chip design projects, a new wave of startups is emerging, opening the door to more affordable custom designs.
These startups are injecting new vitality into the industry, which has favored the largest chip manufacturers due to company consolidation. As large analog chip companies acquire smaller ones, many companies that previously collaborated with other small companies on custom projects can no longer do so due to stricter cooperation terms. This has led to some small customers losing opportunities, but their prospects are changing.
Customization needs
Although digital chip designers often create new features and designs, analog circuits tend to use known functions for the most part. Various parameters of these functions may need to be optimized to improve performance, reduce power consumption, or adapt to new operating environments.
"Sometimes you just need a function or feature that is not available on the market, and then you can make your own ASIC to stand out," said Benjamin Prautsch, head of the Advanced Mixed-Signal Automation Department of the Adaptive Systems Engineering Department at Fraunhofer IIS. "That's the secret to solving this problem."
Advertisement
Analog chip challenges can usually be avoided by using as much digital technology as possible. "By reusing or pushing as much as possible into the digital world, you can minimize the workload," Prautsch explained. "But you will always have an analog shell."Synopsys Director of Product Management Hany Elhak agrees. "If we adopt any digital SoC or microprocessor from Intel, application processors from Qualcomm, or GPUs from NVIDIA, they are primarily digital chips," he says. "But all of these will include specific analog modules."
Due to the highly specialized nature of analog chip design, projects are often outsourced. In the past, these projects were undertaken by well-known analog chip companies, but after being acquired by large companies, these companies may no longer take on such projects—especially for small-batch projects. Now, some projects that acquisition companies are no longer willing to undertake are being taken over by new companies.
Analog chip design remains stubbornly specialized.
In the early days of semiconductor design history, all chips were essentially analog. Even if they performed digital functions, they were designed using typical analog tools and techniques. The small scale of design made this possible. However, at some point, digital design took off in a way that analog design could not achieve.
The key to the breakthrough was abstraction. The simplest is to group certain analog values as a group of digital 1s, and another group as digital 0s (with a protective band between the two groups). "You can abstract the digital as 0s and 1s," says Prautsch. "Beneath it, it is still an analog value, but it can be represented by 0s and 1s. In the analog world, it is always in between."
This abstraction simplified digital design and verification, and its scale is now extraordinary, capable of creating chips equivalent to billions of gates - in the past, these gates might have been carefully crafted by hand. Our current efforts to popularize electronic products everywhere are a direct result of this productivity.
However, analog chip technology has not made such progress. Despite various attempts, analog chip design automation has largely failed because analog technology itself is not suitable for abstraction. "We know the team is studying it, but its level of automation is far lower than what we see in digital technology," observes Prautsch. Single-point tools help address some specific challenges, and they may also help with process node migration, but it is completely different from the digital tool suite.
This has led to some troubles, making analog chip design always challenging. First, analog design still mainly relies on manual work. A module library can be assembled and pieced together, but the boundaries between modules may reduce the performance of fully manual custom designs. And buying IP does not necessarily solve the problem. "Even integrating IP requires analog design expertise," says Elhak.
Second, based on manual work, there are fewer novel analog functions, so analog designers can innovate in other ways. "There is a lot of creativity in improving performance, reducing noise, and increasing speed," explains Elhak. These are usually done one block or one section at a time.
Orca CEO and co-founder Andrew Baker pointed out: "What we call customization is not customization. We are referring to specific applications of standard products customized for specific applications or vertical markets." It is more about optimization rather than a completely new design.In reverse, this also makes analog design more challenging than digital design. While tools can handle many complex issues of large digital chips, the design of analog chips is usually handled by the designers themselves or with simple tools like Excel. Depending on individual personality, the black-and-white characteristics of digital chips may be more attractive to some people than the gray characteristics of analog chips. This means that the community of analog designers is much smaller than that of digital designers. "Students are choosing electronic engineering less and less, and more and more computer science," said Prautsch. "As a result, there is a shortage of new talent for designing analog chips."
The impact of this situation is unexpectedly widespread. Well-funded companies like Google or Amazon, although they have entered the digital design field, still rely on analog contract manufacturers to meet their specialized analog chip needs.
In addition, architects try to use as much digital as possible instead of analog. But the world is analog, so even if 95% of the circuit is digital, most of it must be packaged with analog. This may be because the circuit is interacting with external analog functions, or it just needs to communicate with other digital circuits. For example, I/O physical layer circuits (such as those used for PCIe) are analog.
Baker pointed out: "In most cases, [analog content] accounts for about 20%," most of which is digital content. "Everyone thinks the world is going digital, but analog is still essential in all these systems. Without analog, you have nothing."
Off-the-shelf analog circuits are harder to implement
Digital 1s and 0s make a given digital chip easier to operate in various environments. As long as the signal can be recognized as a high or low level, the circuit will work. Analog circuits are not the case. Although digital circuits may accept 0.8V or 1.0V as a logical high level, these values are very different for analog circuits, and circuits designed to accept up to 1.0V input may not be the best choice for circuits with a maximum input voltage of 0.8V.
Therefore, subtle differences between operating environments can make off-the-shelf chips insufficient (or at least not optimal). For example, chips for standard I/O may not require customization - unless someone needs some enhanced parameters, such as reducing power or improving reliability. "Even within a standard, different companies can achieve better error rates, or they can achieve higher data rates for their own competitive advantage," Elhak pointed out.
Two specific examples of customization needs are Analog-to-Digital Converters (ADC) and Power Management Integrated Circuits (PMIC). ADC is an example where input voltage is important, and the input voltage may vary from system to system. "There is no 12-bit ADC IP that can be suitable for all applications," said Prautsch. "The input voltage is different every time. Sometimes I have a 1.0V single-ended input voltage swing. Sometimes I have a 2.0V differential input voltage swing. Ultimately, it's the same ADC, but it needs a different type of input buffer."
PMICs tend to work based on known or standardized power sources, but these power sources are usually batteries. As battery technology is rapidly developing after years of stagnation, even the standard output voltage of known mechanisms such as lithium-ion technology has changed from about 4.2V to 4.5V, and PMICs designed to handle old standards will not work, especially when controlling battery charging.
"Devices developed ten years ago or even five years ago cannot charge batteries above 4.3 volts," Baker said. "This can hinder their use with newer chemistries. We have improved the accuracy of charge termination, so people can confidently charge the battery to its maximum capacity without overcharging and shortening its lifespan."Some changes may bring more high-end features to more affordable chips. "Customers may be looking for products with slightly lower integration, but high-end features are only available on expensive chips with high integration," Baker said. Dynamic voltage regulation is an example. "We added energy recovery during sharp voltage changes, so when you go from a higher voltage to a lower voltage, the energy stored on the output capacitor is usually just discharged to the ground. We recover this energy back to the input end."
In some cases, such as the battery example, the new product may eventually become a standard. A chip originally customized for a customer may become an off-the-shelf part, such as Orca's PMIC. In other cases, especially those involving patented ideas that others may not be able to adopt, it may remain customized and exclusive. These considerations will affect the cost of obtaining custom analog circuits.
From Analog Circuits to Analog "Systems"
One method used by today's analog chip designers is to use control circuits to adjust various internal parameters in analog circuits. These adjustment points are controlled by digital circuits, which in turn can be controlled by firmware-driven external pins. In this way, a single circuit can operate under a wider range of conditions, so that after power-up, the firmware will set the required configuration.
The combination of analog circuits, digital control, and software drivers is called an analog system. "What we mean by 'system' is that in addition to analog functions, we also add digital control and software," Elhak said. "All these analog modules will have digital circuits that can do everything from selecting oscillator frequencies to adjusting amplifier gains."
Process variability is an increasingly serious problem
Another problem with analog chips is process variability, especially at advanced process nodes. One way to deal with this increasingly serious problem is to calibrate a given chip using adjustment points. In this way, a set of chips (each chip is technically slightly different) can be as close as possible to completely the same to ensure normal operation. "Even traditional analog boxes need to be calibrated by digital circuits and need software control," Elhak said. Since the software driver needs to be the same for each chip, customization is done internally, and eFuse stores the calibration values set during manufacturing testing.
However, this may not be helpful when introducing analog chips to new processes. The most obvious example is introducing them to more modern silicon nodes, but even changing foundries or wafer factories at a given node may require some design adjustments. "Assuming I want to migrate my old design from node 1 to node 2," Elhak explained. "Now I need to optimize it so that it can provide the same performance as the old design, or even better."
This situation can be avoided if the analog function is located on its own chip or chiplet, because analog circuits tend not to benefit from advanced nodes. But if the analog circuit is integrated into an SoC or another digital chip, this situation is inevitable. Even if the chip's function is 100% digital, it still has to communicate with other chips through the lowest level of analog channels, even just to connect to another chiplet. Integrated memory is another example of analog islands in a digital ocean.
Although software-controlled analog chips and IP may reduce the need for customization, they cannot completely eliminate the need for customization. This situation forces chip manufacturers or system manufacturers to make a decision between self-development or purchase. Unless the company has a dedicated analog design team, the decision usually tends to be to buy.Custom Circuit Engagement
The types of custom engagement required are usually less than those needed for fully custom analog chips or circuits. Instead, parts of existing circuits are modified, which narrows the scope of the project. That said, even if the design changes may only involve a few points in the circuit, verification must include the entire circuit because analog is inherently prone to a lot of feedback, so everything can affect each other. "For analog circuits, the situation is much more complex because every change in the circuit can affect the rest," said Elhak.
Therefore, verification must be thorough and may constitute the vast majority of custom work. "Companies use some techniques to reduce the complexity of the problem," he said. "But in the end, they always perform large-scale chip simulation of the entire analog module at the transistor level, and they need to repeat this process in multiple corners and multiple digital control scenarios."
The terms of managing custom projects may vary depending on the specific changes required and the widespread use of these changes beyond a single project. The upfront non-recurring engineering (NRE) costs can be quite high, especially for startups with limited budgets. "This requires a lot of NRE because it requires a lot of manual work," said Prautsch. "You can buy IP, but it often needs to be adjusted. You may not get all the IP you need, or you got the IP, but it's not suitable for the specific semiconductor process you're looking for."
High NRE may be necessary for chips that can sell at a higher price. "If you have an application where the chip manufacturer's customers will squeeze your performance as much as possible, then the price of the product will inevitably be very high. You can only make a few chips with NRE, and in the end, it will pay off because you have solved the problem of expensive niche applications," said Björn Zeugmann, manager of the Integrated Sensor Electronics Department at Fraunhofer IIS/EAS.
It is very attractive to have access to analog chip technology without a large cash investment. If the expected output is high enough and there is a possibility to sell the modified chip or circuit to other customers, the design company may accept a contract without prepayment of NRE. "If a large design company sees a larger market, they may be interested in customization," said Prautsch.
On the other hand, if the customer wants exclusive rights, NRE may be required, and the amount will depend on the scope of the project and the duration and expected quantity of the exclusive rights. If the quantity is too small, then the design company may refuse to undertake the project—especially if the customer cannot afford the NRE costs.
If the chip to be modified has available open design data, the project may become easier, which means that it can be modified by others in addition to the original manufacturer. "Designing a chip for the customer and selling the chip as a black box is an option," said Zeugmann. "Or they can sell open design data so that customers can go to another chip manufacturer or factory or design company for modification."
Changing Design Environment
However, with large companies such as Renesas Electronics, Analog Devices, and NXP acquiring small analog chip companies such as Dialog, Linear Technology, and Maxim Integrated, the prospects for such opportunities have changed recently. These mergers have changed the competitive environment in several ways. The most obvious is that the threshold for custom business has increased five to six times. "You can list a bunch of smaller companies that can seize the opportunity for lifetime revenue of $50 million or $100 million," said Baker. "Now, this threshold has risen to $300 million."The ability to close deals also presupposes that the company wishes for its analog chip team to be involved in such projects. "A large part of these acquisitions in this field is to obtain experienced teams capable of manufacturing specific types of chips, so that you don't have to build your own team," noted Prautsch.
As company cultures blend, conflicts, and coordination processes, as well as deciding which division owns which products, the participation process also becomes more complex. "If I want a customized solution, who should I turn to?" Baker asked. "If someone makes a specific product request to them, there might be three companies within the company's organization willing to help." Until this friction is eliminated, the process of closing deals may be prolonged.
However, new startups are entering the market, trying to fill the gaps left by mergers. Since analog chip design is very specialized and targeted to specific applications, it requires customization and revenue levels below the requirements of large companies. "Consolidation has opened up market opportunities for companies like us to meet the often-heard, unmet needs in the market," Baker said.
In addition to Orca, Elhak also cited CoreHW, Credo, Endura, and Silicon Creations as examples of recent startup activity. "Many of our custom design customers are very small startups, who are either trying to build lidar for autonomous driving or trying to build auxiliary analog circuits for AI accelerators," he said.
Conclusion
Many areas of the semiconductor industry are known for waves of startups and subsequent acquisitions and exits. That being said, the current wave of analog chip startups, following the acquisition wave, may itself become another wave of acquisitions, as a new batch of winners is harvested. But at least for now, the newly small companies will demonstrate the feasibility of analog chip technology at lower revenue levels.
*Disclaimer: This article is the original creation of the author. The content of the article is the author's personal opinion, and our reposting is only for sharing and discussion, does not represent our approval or agreement, if there are any objections, please contact the backstage.
Comments