Intel’s move on the wireless market may fail for the same reason that its X-Scale and ASIC ventures failed – because it’s not putting its wireless parts on advanced processes.
It is only this year, some 18 months after Intel started running its 32nm process, that a wireless chip-set – Medfield – is going to be run on a 32nm process.
The date for putting Atom on the next-generation 22nm process is 2013, although Intel’s 22nm process is due for introduction this year.
Meanwhile, ARM and TSMC are currently making their wireless chips in volume on 28nm silicon and expect to be in 20nm production at the end of this year.
Asked about this, Intel’s Alistair Kemp told me: “Last year we stated our intention of moving Atom up to the front of the process node.” However, the first node at which Atom will be at the front of the node – i.e. one of the first products to be made on that process generation – will be 14nm, says Kemp.
Meanwhile Atom is late to 32nm and late-ish to 22nm assuming that 22nm it gets into production this year.
That is par for the course for an emerging business unit at Intel – new business units have to fight for leading edge capacity and usually don’t get it.
In January 2007, shortly after Intel’s X-Scale business was sold to Marvell, I asked Jeff Krisa, who was general manager of the X-Scale group at Intel, and who went with the group when it was sold and headed the group at Marvell, if one of the reasons for the group’s failure at Intel was because the group was not given access to Intel’s latest processes, Krisa replied: "Access to the latest processes was delayed."
One reason why it was delayed is that Intel’s fab managers have a lot of autonomy over what silicon they run and they prefer to run high value silicon rather than low value silicon.
This process is described by Intel’s former Chairman and CEO Andy Grove In his book, Only the Paranoid Survive. Grove describes how, while he and Gordon Moore were having the discussion which ended with them exiting the DRAM business, Intel employees were already making the decision for them. Grove writes: “Men and women lower in the organisation, unbeknownst to us, got us ready to execute the strategic turn that saved our necks. Over time more and more of our production resources were directed to the emerging microprocessor business, not as a result of any specific strategic direction by senior management, but as a result of daily decisions by middle managers. Bit by bit they allocated more and more of our silicon wafer production capacities to those lines which were more profitable, like microprocessors, by taking production capacity away from the money-losing memory business. By the time we made the decision to exit the memory business, only one out of
So it could very well be that Intel’s fab managers don’t want to run little bits of silicon selling for $20, when they could be running big chunks of silicon which can be sold for several hundred dollars in gazillions of units.
Another reason for the delay in putting its wirless chips onto the latest processes could be because Intel may be finding it is taking longer than it expected to wrap the IP around the Atom core to make a wireless SoC,.
