Back in the 1990s in the middle of the 2G GSM era, cell-phone manufacturers would display a “triangle of difficulty” with a large base labeled radio, a middle smaller part labeled baseband and a little triangle on top labelled software. The idea was that the radio was incredibly difficult, then the baseband chip and there wasn’t a lot of work on software since there wasn’t that much in a phone of that era. But they would point out that that was then and now the triangle was inverted. Radio was basically a solved problem, baseband was still a challenge but more and more manpower was being consumed by software. It was true. The radios didn’t change much, a new baseband at each process node with different speed/power tradeoffs, but more and more of the differentiation was moving into software. Plus nobody was yet trying to integrate the radio onto the pure-CMOS baseband chip, that challenge lay in the future.
But that was then. Then came 3G and, not that long after, LTE. Radio went back to being an unsolved problem. Building the radio interface, also known as the modem, was right on the limit of what was possible given the silicon performance. To make things worse, the power budget could not go up much even though the computational load was much higher.
Qualcomm had distinguished themselves in the 1990s by commercializing CDMA but their claim to success in that era was not that they were better at designing modems than anyone else. Then suddenly they were ahead by a long way. The Gobi series of modems started by supporting both CDMA and GSM on the same modem, along with GPRS and EDGE (data standards).
Several teams tried to design LTE modems. When ST-Ericsson was shut down, its LTE modem was one of the few things of value and was kept by Ericsson in the divorce. Intel needed an LTE modem and it had one in design with the Infineon acquisition, and then went and got another one through an acquisition of Fujitsu’s LTE team. As an indication of how difficult it is, Intel has not managed to get its LTE modem onto its own process and has to manufacture it via TSMC. Almost all the other LTE modems were late.
The current version of Qualcomm’s Gobi handles LTE with speeds up to 300 Mb/s and backward compatibility to 3G, EV-DO (data), and GPS.
Initially LTE modems were standalone chips. For some applications that is still desirable. Apple designs its own application processors (A4…A7) but it uses a Qualcomm Gobi modem as a separate chip. One theory is that designing an LTE modem is too hard even for Apple (although with their cash they could buy a modem design team or a whole company). The other is that getting approved by all the different operators is something they would rather leave to Qualcomm who have to do it anyway and if they integrated the modem on the Ax chips they would have to go through that certification process separately. With a separate chip, modem certification goes on in parallel with application processor design.
Very early on, Qualcomm had the Gobi modem integrated onto its own Snapdragon processors as a single chip. They were way out in front doing this. nVidia is just getting there now. Broadcom (after acquiring Renasas LTE line) now has an integrated product. Intel still isn’t there and won’t be any time soon. Mediatek still has separate chips but is apparently sampling an integrated product.
The result: Qualcomm is far and away the market leader in baseband application processor chips (at least the merchant market, both Apple and Samsung create their own). In LTE-based baseband chips, analysts give it 95% market share. After all the other guys didn’t have anything when it was needed and only now (literally, there are daily announcements from MWC in Barcelona this week) are announcing handsets incorporating their products.
Presumably the handset manufacturers will want to make at least one or two of these competitors viable, for security of supply and price negotiation. So Qualcomm’s business will probably grow but the percentage share will probably shrink. But Qualcomm already have Snapdragon sampling in TSMC’s 20nm process, out ahead of everyone else again. Correction: it looks like just the Gobi modem is in 20nm, not yet Snapdragon (although the modem is far harder to port than a pure digital SoC so it shouldn’t be far behind).
Even the competitors are not bullish. As a senior marketing manager of nVidia said:“We are never going to be at the level of Qualcomm. We have to take baby steps.”
Recently Qualcomm announced Octocore versions of Snapdragon with 64 bit processors and a specialized chip for Automotive built in 20nm.
In the 1990s, Qualcomm’s secret weapon was that they understood CDMA better than anyone else, having invented it. In the last 7 or 8 years it has been their Gobi modem technology, first in 3G but especially for LTE along with the fact that they have had a single chip integrated application processor and modem. They have out-executed the rest of the industry.
As always, mobile is a fascinating industry to watch.
Share this post via: