De-Mystifying The SoC Supply Chain

It’s all about mitigating risk, but that’s becoming harder as complexity increases.

popularity

By Barbara Jorgensen
At the heart of every supply chain operation is the desire to mitigate risk. In theory, a supply chain allows a customer to leverage the best of the best in technology, logistics or production at a lower cost than DIY (do it yourself.) The system on chip (SoC) supply chain is no different—there’s a whole ecosystem in the semiconductor industry that supports design, production, assembly and test. Cobbling together a supply chain isn’t a problem. Technology is.

The SoC supply chain has become as complex as the chips themselves. As semiconductor technologies advance toward smaller geometries and higher levels of integration, customers face an increasing level of risk every time they contemplate a “make” versus “buy” decision. There are problems with integration, new technologies, stacked die and advanced geometries. Customers, as always have two choices: vet, select, engage and manage an array of vendors, or outsource that process to someone else.

“Customers look to us to ‘de-risk’ things,” says Patrick Soheili, vice president for business development at eSilicon. “We have relationships with multiple foundries, IP and service providers, EDA suppliers, and packaging, assembly and test operations. We leverage these relationships to the benefit of the customer. Our goal is to bring a product to market on time, on budget and beyond customer expectations.”

He’s not alone in this assessment. “An IC designer really has to trust a provider of third-party IP,” says Herb Reiter, president of eda2asic Consulting. “There are many providers and many specialize in a certain field, but they need to build a reputation and be reliable. Don’t buy it unless you can trust it.”

Semiconductor services companies (eSilicon uses the term value-chain producers, or VPC) can reduce customer risk at many levels, but their highest value proposition is perhaps managing the unknown. In the design realm, this means IP. VPCs work with EDA vendors, develop IP and contract with third-party IP providers. Selecting the best IP for an application—and then ‘proving’ it—heads off a lot of problems down the line.

Over the past decade, says Huzefa Cutlerywala, senior director of technical solutions for Open-Silicon, his company has vetted a lot of IP. “We’ve developed a dedicated IP team, so even before a customer buys IP, we’ve qualified the vendor,” he says. As chips move from design to first silicon, cost and complexity only become bigger factors. “If you have to respin a mask, for instance, that becomes very expensive,” Cutlerywala says.

With unproven technologies, VCPs say, being transparent with the customer is key. “The process really starts off by understanding your customer’s business,” Cutlerywala says. “In cases in which a customer wants something that’s new or hasn’t been vetted, we’ll explain the risks to the customer upfront. When you work with the right vendors and your customer understands the risks, it usually works out well.”

“When you’re working with smaller geometries, communicating with the customer is imperative,” adds Soheili. “How we manage all of these issues— tools, methodologies, the right libraries—that’s our value. We de-mystify things as much as possible.”

On the production end of the equation, risk—and expense—takes the form of inventory. Proprietary products, such as SoCs, can’t be re-used, resold, or returned. Overestimating a production run could be a disaster, especially for a small or start-up company.

“We understand that if a customer gives us a forecast we will verify the validity of that forecast,” Cutlerywala says. “Some of our customers tend to be a little aggressive, and we have to make sure customers understand there are ramp-up, reliability testing and lead times. It’s a learning curve. But once we set reasonable expectations, moving into mass production becomes a lot easier.”

That kind of information is also valuable for the foundry, Reiter says. “It’s very difficult to manage production flow when you are signed up in joint development with five, 10 or more customers. The inventory risk is a liability if things go wrong.” Contracts go a long way toward protecting all the parties involved, but they don’t replace constant communication. “By organizing the design flow; corporations having a lot of back-and-forth communication and mapping out the process, you’re basically distributing the risk—and the reward.”

It’s rare that production ever goes smoothly. “When there are mishaps and a customer has to cancel, there are contractual clauses for that,” Soheili says. “But we never want to reach that point. We try to do our best in negotiating [with the foundries].” Most customers, he says, realize bringing a chip to market is long and drawn-out process. “If we’re working on a 28nm chip, customers know the first time they’ll see some meaningful revenue is four years down the road. There are a lot of risks there to begin with.”

Ultimately, any supply chain is designed to manage cost, but partners are in business to make a profit. Managing these relationships can be tricky, especially when there are multiple parties. “When you have three or four layers of people and every one of them adds a margin, it’s called margin stacking,” explains Soheili. “That’s not the business we are in. Our job is to build a chip for less than it would cost you to build it yourself. eSilicon maintains relationships with TSMC, other foundries and test and assembly houses for that reason. Our value is not just putting the partnerships together or being a legal clearinghouse—our job is to assemble a chip that’s smaller, runs a watt and a half lower, or faster. If we can’t do that, we shouldn’t be in the business.”

There are some tools to help with all of this, as well. Mentor Graphics and Dassault Systemes have developed product lifecycle management software. In Mentor’s case, that software has been used for everything from tracking costs to real-time swapping out of components that can be used on a PCB or in an automotive bill of materials. Dassault takes a somewhat different approach, looking at the global supply chain using a 3D model. And both companies integrate a large number of other tools with their PLM offerings, which can help control costs across the supply chain even for companies that only play in part of it.