Analog Consolidation Spurs New Round Of Startups

A new wave of startups is rising to meet the growing need for specialized analog customization in chip design projects, opening the door to more affordable custom designs.

These startups are breathing new life into a sector, which as a result of consolidation has favored only the largest chipmakers. As larger analog companies acquire smaller ones, many companies that were previously engaged in customization projects with other small firms are no longer able to do so due to more restrictive terms of engagement. This has left some smaller customers out of luck, but their prospects are changing.

The need for customization
While digital designers create novel functions and designs regularly, analog circuits largely tend to stick with known functions. Various parameters for those functions may require optimization to raise performance, lower power, or adapt to a new operating environment.

“Sometimes you simply need a feature or functionality that’s not available in the market, and you make your own ASIC to differentiate,” said Benjamin Prautsch, group manager for advanced mixed-signal automation at Fraunhofer IIS’ Engineering of Adaptive Systems Division. “That’s your secret sauce to address this niche.”

Analog challenges often are avoided by using digital techniques as much as possible. “You reduce the effort as much as possible by re-using stuff or pushing as much into the digital world as possible,” explained Prautsch. “But you always will have an analog shell.”

Hany Elhak, director, product management at Synopsys, agreed. “If we take any digital SoC or a microprocessor from Intel, or an app processor from Qualcomm, or a GPU from NVIDIA, they’re predominantly digital chips,” he said. “But all will include specific analog blocks.”

Because analog design is such a specialized area, projects typically are contracted out. In the past, they were taken on by well-known analog houses, but after being merged into larger companies, those companies may no longer be available — particularly for smaller volumes. New companies are now picking up some of the projects that the acquiring companies are no longer willing to entertain.

Analog design remains stubbornly specialized
Early in the history of semiconductor design, all chips were effectively analog. Even if they performed digital functions, they were designed using typical analog tools and techniques. The small scale of the designs made this possible. At some point, however, digital design took off in a way that hasn’t been possible with analog.

The key to that breakout was abstraction. The simplest was the grouping together of certain analog values as a digital 1 and the other group as a digital 0 (with guardband between). “You can abstract digital to 0s and 1s,” said Prautsch. “Below, it’s still analog, but it can be represented as 0s and 1s. In the analog world, it’s always everything in between.”

That abstraction simplified digital design and verification, and the scale is now extraordinary, creating chips with the equivalent of billions of gates — each of which might have been carefully hand-crafted in the past. Our current efforts to put electronics everywhere are a direct result of that productivity.

Analog, however, has seen little such progress. Despite various attempts, analog design automation has largely failed because analog doesn’t lend itself well to abstraction. “We know that crews are working on it, but it lacks by far the level of automation that we see in digital,” observed Prautsch. Point tools help with some specific challenges, and they also may help with process-node migration, but it’s nothing like the suite of digital tools.

This leads to several impacts that keeps analog design challenging. First, analog design is still largely manual. It’s possible to assemble a library of building blocks and stitch them together, but the boundaries between blocks may dial back performance from what would be possible with a fully manual custom design. And buying IP doesn’t necessarily solve the problem. “Even integrating IP requires analog design expertise,” said Elhak.

Second, based on manual efforts, there is less in the way of novel analog functionality, so analog designers are creative in other ways. “There is a lot of creativity in improving performance, reducing noise, increasing speed, things like that,” explained Elhak. That often is done a block or a segment at a time.

“By customized, we don’t mean custom,” noted Andrew Baker, CEO and co-founder at Orca. “We mean application-specific standard products that have been customized for a specific application or market vertical.” It’s optimization more than it is fresh design.

This, in turn, makes analog design more challenging than digital design. While tools can handle many of the intricacies of large digital chips, their analog equivalents are often managed in the designers’ heads, or with simple tools such as Excel. Depending on one’s disposition, digital’s black-and-white nature may appeal more to some people than the shades-of-gray nature of analog. This means the community of analog designers is far smaller than that of their digital counterparts. “Students are going into electronics engineering less, and more into computer science,” said Prautsch. “As a result, there is a shortage of fresh new personnel designing analog chips.”

The situation has surprisingly broad impact, with deep-pocketed companies such as Google or Amazon — which have taken on digital design — still relying on analog contract houses for specialized analog requirements.

Further, architects attempt to implement as much as possible in digital rather than analog. But the world is analog, so even if 95% of a circuit is digital, much of it must be wrapped in analog. That may be because the circuit is interacting with an external analog function or just needs to communicate with other digital circuits. For example, I/O physical-layer circuits, such as those used for PCIe, are analog.

“In most of our cases, [the analog content] is going to be in the 20% range,” with the bulk being digital, Baker noted. “Everybody thinks that the world is going digital, but analog is still essential in all these systems to enable them. Without analog, you’ve got nothing.”

Off-the-shelf is harder for analog
Digital 1s and 0s make it easier for a given digital chip to operate in a variety of environments. As long as a signal can be recognized as being high or low, the circuits will work. This is not the case with analog. While a digital circuit might accept 0.8 V or 1.0 V as a logic high, those are very different values for analog, and a circuit designed to accept up to 1.0 V on an input may be sub-optimal for a circuit with a maximum input voltage of 0.8 V.

As a result, subtle differences between operating environments may make off-the-shelf chips inadequate (or at least sub-optimal). Those used for standard I/Os, for example, may not require customization — unless someone has a need for some enhanced parameter such as lower power or a higher reliability. “Even within a standard, different companies can achieve a better bit error rate, or they can get a higher data rate for their own competitive advantage,” noted Elhak.

Two specific examples of the need to customize are analog-to-digital converters (ADCs) and power management ICs (PMICs). ADCs are an example where the input voltage matters, and it can vary from system to system. “There is not one, say, 12-bit ADC IP that is universal for every application,” said Prautsch. “The input voltage differs every time. Sometimes I have a 1.0 V input voltage swing, single-ended. Sometimes I have a 2.0 V differential. It’s the same ADC in the end, but it needs a different kind of input buffer.”

PMICs tend to work based on known or standardized power sources, but those often can be batteries. And because battery technology is evolving at a rapid clip after remaining stagnant for years, even known mechanisms like lithium-ion technology are seeing the standard output voltage move from about 4.2 V to 4.5 V, and PMICs designed to handle the older standard will not work, particularly when controlling the recharging of the battery.

“Devices developed a decade or even five years ago don’t have the ability to charge a battery that’s above 4.3 volts,” said Baker. “That can preclude its use with newer chemistries. We’ve improve the accuracy of the charge termination so one can confidently charge to the maximum capacity of the battery without overcharging it and reducing its lifetime.”

Some changes may bring higher-end capabilities to chips with more affordable prices. “Customers may be looking for something with slightly lower integration, but with high-end features available only in highly integrated expensive chips,” said Baker. Dynamic voltage scaling is an example. “We’ve added recycled energy during aggressive changes in voltage, so when you go from a higher voltage to a lower voltage, the energy stored on the output capacitors is typically just discharged to ground. We recycle that energy back to the input side.”

In some cases, such as the battery example, what’s new eventually may become standard. A chip that might start out as a custom alteration for one customer may become an off-the-shelf part, such as Orca’s PMIC. Other cases, especially those involving patented ideas that others may not be able to take on, could remain custom and exclusive. These considerations affect the cost of obtaining a custom analog circuit.

From analog circuits to analog “systems”
One approach used by analog designers today involves instrumenting an analog circuit with control circuits to adjust various internal parameters. Those adjustment points are controlled by digital circuits, which in turn can be controlled through external pins driven by firmware. In this manner, a single circuit can operate in a broader set of cases such that, after powering up, firmware sets the desired configuration.

The combination of analog circuit, digital control, and software drivers has been referred to as an analog system. “By ‘systems,’ we say that, in addition to the analog function, we are adding digital control and we are adding software,” said Elhak. “All these analog blocks will have digital circuits that accomplish, for example, anything from choosing the frequency of an oscillator to adjusting the gain of an amplifier.”

Process variability a worsening problem
An additional problem for analog is process variability, especially at advanced process nodes. One way of handling this worsening problem is to calibrate a given chip with adjustment points. This allows a complete family of chips, each of which technically operates slightly differently, to behave as close to identically as necessary for correct functioning. “Even traditional analog boxes still need to be calibrated by digital circuits and need to be software-controlled,” said Elhak. Because the software driver would need to be the same for each chip, the customization happens internally, with eFuses storing calibration values set during manufacturing test.

However, this may not help when taking an analog chip onto a new process. The most obvious example of this is taking it into a more modern silicon node, but even changing foundries or fabs on a given node may require some design adjustments. “Let’s say I want to migrate my older design from Node 1 to Node 2,” explained Elhak. “Now I need to optimize it so that it gives me the same performance as the older one, if not better.”

Such a situation may be avoidable where the analog function is on its own chip or chiplet, because analog circuits tend not to benefit from advanced nodes. But it can’t be avoided if the analog circuit is to be integrated onto an SoC or another digital chip. Even if the chip’s function is 100% digital, it still must communicate with other chips over channels that are, at the lowest level, analog, even if just to connect to another chiplet. Integrated memory is another example of an analog island in a sea of digital.

While software-controlled analog chips and IP may reduce the need for customization, they don’t eliminate it entirely. Such a situation gives a chipmaker or system-maker a make-or-buy decision. And unless the company has a dedicated analog design team, that decision typically leans toward buying.

Customized circuit engagements
The types of customization engagements needed typically fall short of full-custom analog chips or circuits. Instead, portions of an existing circuit are modified, which reduces the scope of the project. That said, even though design changes may touch only a few spots in the circuit, verification must include the entire circuit because analog, by its very nature, tends to involve a lot of feedback, and so everything conceivably talks to everything. “With analog circuits, it’s a lot more complex in that every changing part of the circuit impacts the other parts,” said Elhak.

Therefore, verification must be thorough and may constitute a big part of the customization effort. “There are techniques that companies use to reduce the complexity of the problem,” he said. “But in the end, they will always end up running a very big simulation at the transistor level for the full analog block, and they will need to repeat that for multiple corners and for multiple digital control scenarios.”

The terms governing a customization project likely will vary based on the specific changes required and how broadly they might be useful beyond the one project. Up-front non-recurring engineering (NRE) charges can be substantial, particularly for startups with limited budgets. “It’s a lot of NRE, because it’s a lot of manual work,” said Prautsch. “You can buy IP, but it must often be adapted. You probably won’t get all the IP that you need, or maybe you get the IP, but it’s not available in the specific semiconductor process you’re looking for.”

A high NRE may be necessary for chips that can command a higher price. “If you have an application where the customer of the chipmaker is going to squeeze out whatever performance is possible, then obviously the price of that product can be really high. You can build just a few chips with an NRE where, in the end, it will pay off just because you exactly address the expensive niche application,” said Björn Zeugmann, group manager for integrated sensor electronics at Fraunhofer IIS/EAS.

The ability to acquire analog technology without a big cash hit upfront can be attractive. Design houses may take on such a contract with no up-front NRE if projected volumes are high enough and there’s a potential to sell the modified chip or circuit to other customers. “The big design houses may be interested in making customizations if they see a bigger resulting market,” said Prautsch.

If the customer wishes exclusivity, on the other hand, an NRE may be necessary, and the amount will depend on the project scope as well as the term of exclusivity and anticipated volumes. If volumes are too low, then a design house may decline to take on the project — particularly if the customer can’t afford NRE charges.

The project may become easier if a given chip being modified has open design data available, meaning that someone other than the original manufacturer could modify it. “Designing a chip for a customer and selling the chip as black box is one option,” said Zeugmann. “Or they can sell the open design data so that the customer can go to another chip manufacturer or factory or design house to do modifications.”

A changing design environment
The landscape for such opportunities has changed recently, however, with the acquisition of smaller analog houses such as Dialog, Linear Technology, and Maxim Integrated by bigger companies such as Renesas, Analog Devices, and NXP. These mergers alter the playing field in a couple of ways. The most obvious is that the bar for a customization engagement has moved up as much as five- or sixfold. “You can list a whole bunch of smaller companies that would address opportunities of $50 million or $100 million worth of lifetime revenue,” said Baker. “Now that bar has gone to $300 million in order to take a look at it.”

The ability to strike a deal also assumes that the company wishes to make its analog team available for such projects. “A big part of these acquisitions in this field is to get experienced teams that are able to build a specific type of chip so that you don’t have to build your own team,” noted Prautsch.

The engagement process also becomes more complicated as company cultures merge, clash, and reconcile processes and decide which divisions own which products. “If I want a customized solution, whom do I go to?” asked Baker. “There might be three businesses within the corporate organization that may hold up their hand if somebody comes to them with a particular product requirement.” Until such friction smooths out, the process of getting a deal may be extended.

New startups are entering the market, however, and they are trying to fill the very void that the mergers have left. Because analog design is so specialized and specific to an application, it leaves a demand for customization at revenue levels below those required by larger companies. “Consolidation has opened up opportunities in the market for companies like us to address an underserved need within the market that we hear commonly from customers,” said Baker.

In addition to Orca, Elhak cited CoreHW, Credo, Endura, and Silicon Creations as examples of recent start up activity. “Many of our custom design customers are very small startups who are either trying to build lidar for self-driving, or they’re trying to build accessory analog circuits for AI accelerators,” he said.

Conclusion
Many corners of the semiconductor industry are famous for waves of startups followed by waves of acquisitions and dropouts. That said, the current wave of analog startups, which has followed a wave of buy-ups, may itself become another acquisition wave as the winners of the latest crop are harvested again. But for now, at least, the latest small companies will be around to demonstrate analog’s viability at lower revenue levels.

Related Reading
Integration Hurdles For Analog And RF In Next-Gen Packages
Strategies and advancements for thermal dissipation, shielding, and testing schemes.
Startup Funding: Q2 2024
Big rounds for AI chips; 91 startups raise $2.6 billion.