The first pint of Laughing Lab for the weekend should be tipped in the direction of Rick Merritt of EE Times, who provided us with a hilarious yet quite serious observation on why designers are not likely to consider the Intel Atom in a system-on-a-chip (SoC) design. His six reasons all have merit (pun not intended), such as high royalties, business-term gotchas, and little visibility into architectural details.
But Merritt’s final point, that Intel does not know how to sell processor cores, should be expanded to the embedded world in general. Intel executives, at the highest levels, have not shown the commitment to license products to system vendors, nor have they shown much sense in working with semiconductor partners to provide end-to-end embedded solutions.
In the 1990s, Intel had one of the coolest RISC processors for the likes of printers and network equipment, the i960. The processor had dozens of customers, yet Intel deliberately cut off support under the belief that any business distracting Intel from the desktop and data center should not be tolerated. Today, Intel has to diversify into smartphones, netbooks, and other platforms representing new territory, yet it has not demonstrated a good ability to play well with others.
Certainly, conditions are not always rosy for ARM as it expands into coprocessor cores, as Peter Clarke pointed out in another recent EE Times piece. But ARM, MIPS, and even the likes of Tensilica or Wintegra, have distinct advantages in pursuing big, integrated chip designs. They know how to license cores cost-effectively, and they have no standalone high-volume chip business to defend. Until Intel learns how to turn Atom into a multivendor “Ecosystem” (forgive me, I hate that word), its Atom will really be a non-player outside the basic Intel single-chip designs.
Loring
