Proprietary Memory: A High-Risk Endeavor Leave a comment

//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>

Semiconductor technologies live and die by industry standards, but are there times when it makes sense to build a heavily customized—even proprietary—memory device?

The chip sector is replete with standards organizations that guide the evolution of widely adopted memory devices. JEDEC is responsible for DRAM, LPPDR, GDDR and high-bandwidth memory (HBM), among others. The Peripheral Component Interconnect Express Special Interest Group takes care of the most ubiquitous protocol for data movement, while NVMe Express and the CXL Consortium built their specifications with PCIe as their foundation. Most recently, the UCIe was developed to bring best practices to chiplets.

Graham Allan, senior manager of product marketing at Synopsys, has been in the DRAM business since 1985. “The highway of DRAM technologies is littered with the roadkill of non-JEDEC-standard memories,” he said. “If something isn’t JEDEC-standard, or if a DRAM vendor tries to go it alone with something that they want to differentiate with, it’s going to die.”

Among the many abandoned memories that never saw widespread use are Micron Technology’s HBM competitor, Hybrid Memory Cube, and Rambus’s virtual-channel DRAM, even though the latter had the backing of Intel. “These things all die because they don’t have the benefit of the entire ecosystem,” Allan said.

He said the value of a standardized memory device is that it can be multi-sourced—an SK Hynix product will plug into the same socket as a Micron product.

That’s not to say some vendors aren’t offering special features, Allan noted. “There will be DRAM vendors that will have their own special features that they unlock for special customers, and it’s still standard DRAM,” he said. “If you have the technology under the hood to take advantage of this little extra stuff that they put in there, that’s going to make the product all that much better.”

Sole-source scenarios put vendors in a corner

Some memories that are developed by more than one vendor may show potential but are subsequently abandoned by all but one of the vendors. Reduced-latency DRAM was initially developed by Infineon Technologies in the late 1990s; Micron was subsequently brought in as a development partner and a second source, but Infineon opted to exit the market.

Allan said the challenge of being the sole source of a memory product that has a strong customer base—especially if it’s a Tier 1 customer—is that it’s difficult to end-of-life the product. “When you get a few key design wins for something that’s a niche, custom product that’s not standard, you get sucked into that vortex,” he said.

The danger to a vendor of being the only one to sell a memory device is that you risk getting blacklisted by that customer for a long time if they decide to discontinue it, Allan added.

Jim Handy, principal analyst at Objective Analysis, said that there’s a lot of characteristics that can classify a memory device as proprietary—a DRAM that is wide and narrow to meet a customer requirement, for example. “That’s not uncommon,” he said.

How a memory is packaged can make it proprietary, even if it’s a standard chip inside, Handy said, because it’s being used in a harsh environment, such as deep-sea exploration.

Blurred lines between custom and proprietary

Handy noted that HBM started out as somewhat proprietary—it was a memory that SK Hynix made for AMD. He said this doesn’t happen often because one of the focal points of memories as a business is that they are identical to each other. “That ends up making them a lot cheaper, because instead of having something that’s only being bought by AMD, you’ve got something that’s being bought by AMD and 170 other companies,” he said.

Handy said it’s important to distinguish between “custom” and “sole source.” When it’s developed to customer’s requests, it’s a custom memory; sole source is when the manufacturer is the only manufacturer of the product.

Either path is going to be more expensive, and ideally, being able to buy commercial off-the-shelf components is preferred, even for specialized applications found in military and aerospace sectors, because using a standard product is cheaper than a custom chip.

Standard memories inspire an engineering ecosystem to be built around it, which means customers have less risk—if their preferred supplier exits the market for whatever reason, they have other options.

Intel’s 3D Xpoint–based Optane technology required a great deal of engineering to implement. “Intel spent a lot of money trying to make that happen,” Handy said. Micron knew the technology wasn’t profitable for Intel, which is why it never went to market with a 3D Xpoint SSD or DIMM. “If there were a big enough market, Samsung would have reverse-engineered the product and figured out how to make one of their own.”

Upfront risk can pay off down the road

For Macronix International, an integrated device manufacturer in the nonvolatile memory market, developing a proprietary product for a customer can be worth the risk in the long term if it leads to high volumes and a long-term partnership.

In an exclusive interview with EE Times, Macronix chairman and CEO Miin Wu said he’s ultimately responsible for deciding if it makes sense to provide a proprietary or highly customized product to a customer with specific requirements. Aside from the financial reward if the upfront investment pays off, the work itself can be quite rewarding for those working on a new product line.

Wu said some applications and market segments are more likely to have custom requirements, which Macronix can solve, such as gaming and security. Working with a customer or partner on a custom design may not immediately appear to be a high-volume business opportunity, but it can open the door to future opportunities.

Macronix’s ArmorFlash technology is an example of a customized product designed for a specific customer opportunity that gave birth to a wider product category because it addresses common security challenges as more devices connect to the internet and are pushed out to the network edge.

Wu said Macronix is focused on developing new solutions in the marketplace rather than just being another supplier competing with other vendors selling similar products. “We are creating our own product; we are creating our value,” he said.

Any custom product that Macronix makes must meet high-quality standards, Wu said, and there needs to be strong indications that it will become a high-volume business. Certain applications raise the likelihood that there are long-term benefits for taking a risk on a custom or proprietary product, such as security or aerospace, that will lead to broader market opportunities, he said, especially in industries that require a high degree of qualification for harsh environments with extreme temperatures.

Unique projects drive innovation, skills development

Partnerships with other companies, such as Macronix’s relationship with Nvidia, are examples of how getting onboard early with a developing market can lead to big payoffs—Wu noted that Nvidia has essentially created a new market.

Taking the road less traveled also allows Macronix and its employees to develop new skillsets that can be applied to all parts of the business. “If we create a project for our engineers to challenge themselves, eventually that will really help us to create protections in our future business,” Wu said.

Like Macronix, Phison has no interest in being a “fast follower,” but the company is discerning when it comes to developing niche or proprietary products.

In an exclusive interview with EE Times, Phison CTO Sebastien Jean said everything has caveats when it comes to developing proprietary or heavily customized solutions for customers—and the line between them is blurry. “If you want to actually customize low-level signaling, forget it,” he said. “No one will make something that’s interoperable with your device unless you’re Samsung and you just make all the parts yourself and you’re selling to yourself.”

Jean said making devices that go against standards by swapping signals and pins can literally cause sparks—creating something proprietary is not worth the trouble because nothing ends up working together and customers get frustrated.

An example of a truly proprietary storage device was Sony’s short-lived memory stick, Jean said, which had a unique form factor that looked a lot like an SD card that was squished down and made longer. There were no obvious benefits, he said. “They were doing that to lock people into a form factor so they would only buy them from a Sony. Kind of like printer ink cartridges.”

Customer vetting is critical

Customization that leverages standards-based protocols does have value, Jean said, but it raises issues around intellectual property (IP): “There’s a legal distinction.” He said customers coming to Phison knowing what changes they want to own the IP, but if Phison comes up with the solution, there’s an opportunity to apply it to other customers, even if it can’t disclose details of the original customer engagement that sparked the idea.

The potential for high volumes is a consideration for Phison, but so is the confidence that the customer will be around for the long run, Jean said.

But while many big players will immediately turn down low-volume work, Phison will consider it by vetting the customer and will ultimately charge for any non-recurring engineering (NRE), Jean said. “It weeds out the people who ask for stuff and have no real plan to actually make it real or sell it, because ultimately, we don’t make money off of the NRE,” he said. “It covers the opportunity costs and it ensures engagement.”

These projects need to have executive sponsorship from the customer to demonstrate they are serious, Jean said. “Otherwise, it will very likely die on the vine.” He said if a customer comes and asks for something and they’re willing to pay for it, Phison will take it on if its engineering resources are available and not tied up somewhere else.

Even if the work does go ahead, it is ultimately up to the customer to sell its product, Jean said.

The benefit of taking on custom projects, especially from big players, can be valuable in a tight-knit industry, he added, as word gets around that Phison is working with Tier 1 companies, which is good word of mouth and the company’s reputation.

Ultimately, Phison must remain aggressive on its technology path, Jean said. “We also have to be perceived as technology innovators.”