Did you hear that? It sounded like a couple of big semiconductor firms left a part of the market and a couple of smaller fabless companies stepped in to try to make money with two popular embedded processor architectures.
Two major semiconductor vendors recently announced plans to sell off their well-known but financially underperforming embedded processor families:
AMD sold their Alchemy line of processors to Raza Microelectronics, Inc. (RMI)
Intel sold their XScale line of processors to Marvell Semiconductor
Both these architectures were acquired just a few short years ago when AMD and Intel thought handheld devices would be their next great battleground. But both companies decided they couldn’t make a sustainable profit on these parts.
Most analysis of these deals has highlighted the benefits AMD and Intel may gain from shedding these pieces. But I see something else going on here – the move to a fabless model for embedded processor firms. Can fabless companies perform better with these architectures?
“Smaller fabless firms are generally more nimble, better focused, and better able to leverage available contract fab technology and capacity.”
RMI gets the magic
RMI’s acquisition of the Alchemy lines seems to be a natural fit for them, as they develop advanced semiconductor solutions for information infrastructure, network security, wireless, and digital consumer markets.
RMI is a relatively small private company with undisclosed revenue figures. RMI’s CEO, Atiq Raza, previously founded NexGen and served as COO of AMD after NexGen was acquired. The small firm focuses on delivering power efficient solutions with their most popular product to date, the XLR processor family, a high-end MIPS64 multicore architecture for network acceleration functions.
The Alchemy product lines create an opportunity for RMI to enter the lower end of the MIPS spectrum in their application space, but they’ll face stiff competition there. The real win for RMI may be an agreement to collaborate on Torrenza, AMD’s ecosystem effort to open up and promote Direct Connect Architecture and HyperTransport technologies around the AMD64 core architecture.
Marvell scales up
Marvell is best known for providing switching, transceiver, wireless, communication controller, and storage silicon. Their acquisition of the XScale lines immediately gives them a strong position in some of today’s latest consumer devices, including the Motorola Q, the Palm Treo 700p and 700w, and the RIM Blackberry 8700.
Marvell had fiscal 2006 revenue of $1.67 billion. Intel will continue to fab the XScale parts until Marvell transitions them into their fab suppliers. Marvell had already been optimizing their WLAN products to work with XScale processors, so the technical synergy of the deal is clear. With their portfolio, focus, and some new innovation, Marvell could create critical mass for XScale devices where Intel struggled.
Ab fabless is money
Is fabless the key to making money in the embedded space? I’d say so, and the acquiring companies in these deals would probably agree.
Big, asset-heavy companies, especially those focused on high-end processors, are often lost when it comes to smaller, lower-cost embedded devices. These firms have giant, multibillion wafer fabs that must be fed by enormous volumes of very large, complex wafers. It’s the operational excellence model applied to high tech – mass produce items where the barriers to entry are high, and profit by creating and executing high-yield processes. This is a key requirement for a device such as an Opteron or a Core Duo, but it’s not quite the right model for less expensive embedded processors.
Smaller fabless firms are generally more nimble, better focused, and better able to leverage available contract fab technology and capacity. This frees them to spend intellectual capital innovating new, creative solutions for embedded markets, while still remaining cost competitive. Other benefits include more concentrated sales and marketing efforts, shorter development cycles, and potential profitability on a smaller volume of shipments.
The fabless semiconductor model fits with this issue’s theme of mass customization, which we think is the wave of the future in Embedded Computing Design. I think Marvell has a good chance to put life back into the XScale architecture, while I see the win for Raza as a larger partnership with AMD. Whatever the outcomes, I consider the fabless model the right way to go. What do you think? Are these deals a good thing for our industry? Do fabless business models seem to work? We welcome your thoughts and ideas. E-mail me at ddingee@opensystems-publishing.com.
If you found this article valuable, click here for your complimentary subscription to Embedded Computing Design magazine. You can also subscribe to the Embedded E-Letter, our free e-mail newsletter, by clicking here.
