Texas Instruments Inc (TXN)

[slacker711]

A pretty good overview of the seven year effort by TI to bring this chip to market....

http://www.eetimes.com/showArticle.jhtml?articleID=57702944

Seven-year odyssey nets one-chip phone
By Patrick Mannion
EE Times
January 24, 2005 (9:00 AM EST)



Manhasset, N.Y. — Being called crazy didn't really bother the ragtag cadre of wireless pilgrims. Indeed, the disparaging term was a badge of honor for a tight-knit group born of its members' willingness to attempt a feat that had been broadly assumed impossible.


Rather, what bothered the foursome during the long, dark nights of their seven-year engineering odyssey were the times they let themselves wonder, however fleetingly, whether their detractors were right.


The journey began innocuously enough in 1997 with a wireless-technology road map by Texas Instruments Inc. The realization quickly followed that reaching the defined targets cost-effectively would require the assignees to do something radical. The upshot was a fundamental redesign of radio architectures that would require the company to bring to bear its vast collection of resources and expertise.


On a superficial level, the company's process technology, systems knowledge, IC design, test and simulation, and manufacturing knowledge were elemental and indispensable to the effort. But the deeper foundation was a company culture centered on technical accomplishment and innovation, where bonuses come second to pride, passion and the desire to stake out one's place in a highly competitive technical hierarchy.


It was that combination of the tangible and intangible that kept Ken Maggio, Dirk Leipold, Bogdan Staszewski and Khurram Muhammad focused on their path toward vindication. It came, finally, in the form of the industry's first single-chip all-CMOS GSM phone, for which Nokia is the first customer.


Guided by TI luminaries and industry veterans Dennis Buss, Gilles Delfassey and Bill Krenik, the team's accomplishment at first blush simply follows through on a highly publicized promise made in September 2002. But for TI and those analysts paying attention, the chip's technical underpinnings, based on digital RF processing (DRP), will translate in the near future to highly integrated communications and entertainment devices with ubiquitous low-cost wireless connectivity.


Initially, that connectivity will derive from the ability to integrate multiple radios cost-effectively on a single CMOS device. The longer-term goal, however, is to leverage the attributes of DRP to realize the wireless Holy Grail: software-defined radio.


"This clearly sets a new integration benchmark in terms of what can be put together on single-chip CMOS," said TI CEO Rich Templeton, who had added corporate focus to the list of intangibles that aided in the design's realization. "You have to have a clear idea of targets in terms of products, markets and customers that you clearly intend to win. You have to make sure everybody inside the company knows what that [vision] is — and that people have the resources to get it done."


That rang true for Muhammad, who was responsible for much of the chip's receive chain. "We were given the freedom to craft whatever we wanted, whatever made sense to us — as well as the responsibility to make it work," he said. "That's why we worked all those countless nights."


Now that DRP has put radios into the digital domain, Templeton is confident the technology will have far broader architectural implications. "I'm respectful of not trying to overstate this, but I think 10 years from now, you guys [EE Times] may look back and say DRP was one of the most significant innovations in the IC industry in the first decade of the 2000s."


Templeton's conviction stems not just from the breakthroughs in digital frequency synthesis and direct RF sampling, but also from the end application. With more than 600 million cell phones shipping annually, Templeton considers phones the most important end-product category in the tech industry. That said, he foresees a role for DRP in the wider communications and entertainment worlds, "since it's about radio communications with highly integrated low-cost CMOS — and we get pretty excited about both of those themes."


While Templeton is cautious about software-defined radio, given its myriad definitions and 20-year history of underachievement, he nonetheless thinks SDR's time is at hand. "We're here today," he said. "We have CMOS capable of switching at [the necessary] speeds, and we can integrate it in a low-cost platform."


Max Baron, an analyst at In-Stat/Microprocessor Report, was less reserved. "SDR is where [the technology's] going," he said, though he expressed disappointment in TI's timeline of 2008 to achieve it. Acknowledging that the GSM chip is commercially oriented, he added, "SDR is really the evolution — I'd like to see it happen sooner."


In the meantime, Baron considers the current GSM chip to be critical in terms of component cost and size reduction, as well as reliability. Estimating the cost reduction at "between $1 and $2, tops," Baron said the component cost reduction may not be as significant to customer Nokia as the integration — which itself leads to cost savings in terms of placement, test and inventory management — and the improved reliability. More space will also allow greater functionality, he noted, predicting that TI will quickly add in its Omap and Omap2 line of smart-phone processors.

Misnomer?
Historically, claims of "single chip" wireless have been met with skepticism. That has held particularly true for narrowband systems such as GSM, where receiver sensitivity, output power and phase-noise requirements have confounded integration and derailed attempts at all-digital approaches. That said, the TI chip packs the full digital baseband; all the SRAM, logic and processors; and the entire transceiver — from low-noise amplifiers on the receive side to the power amplifier buffer on the transmit side.


"It includes the VCOs, frequency-generation components, analog functions and power management," said Krenik, wireless advanced-architectures manager at TI. Targeting low-cost, high-volume GSM voice-centric designs, the chip's radio technology is fully GPRS-capable, said Krenik. Validation of that claim is in process. "We're also working on Edge and UMTS [3G]," he said.


With a nod to honesty under the cold eye of industry, Krenik also listed portions that are not on-chip, such as the power amplifier (PA), program memory (flash) and some battery management and power control, as well as some front-end SAW filters and a transmit/receive switch. "But it still [integrates] far and away the lion's share of the circuitry associated with the cell phone," he said.


The decision to leave some of the battery management and power control off-chip had to do with the high voltages involved as well as system-partitioning concerns, Krenik said. "The partitioning was done for cost reasons and for our own road map," he said. "It was designed to have specific advantages at the system level" — not because it would make for "a nice press release."


In-Stat's Baron found no fault either with the partitioning or with the decision to leave the power amp off-chip. "It makes sense to have the PA separate," he said. "Aside from its being difficult to integrate, there are lots of tricks [that can be used] to differentiate it."


Power management, meanwhile, is a system-wide issue and thus is not fodder for complete integration, Baron said. "If you have it inside, it's probably going to have to be super-programmable and complex." Off-chip power management, he said, "allows the cell phone manufacturer to control how they're using power."


The chip started sampling to Nokia in December, so TI fulfilled its September 2002 promise to sample a single-chip phone sometime in 2004. The company expects the chip to ship in production volumes next year.

Big berg
The chip itself is just the tip of a very large iceberg that broke off TI's wireless strategy. Shortly after the road map was released in 1997, it became clear within TI "that we had to do something very radical to be cost-effective and competitive," said Leipold, a physicist from Freising, Germany, who moved to Dallas to join Maggio on the project.


Dallas staffers Muhammad, born in Pakistan, and Polish-born Staszewski later joined the team's co-founders to form the foursome that Krenik calls the fathers of DRP. The team grew over time to include staffers in India and Nice, France. Over the course of the project, team members logged 1,500 hours on project-specific conference calls, exchanged 91,000 e-mails, logged 3 million miles in the air and celebrated the births of 10 babies.


It was Leipold who first made the case for building the chip in CMOS and who said the 90-nm process node would be the jump-in point. But doing analog and RF design at deep-submicron nodes raises seemingly insurmountable problems. And though the four core teammates were accomplished in their respective fields, none had experience in RF design. All, except for Leipold, had come from TI's storage division, where they had focused on hard drive read channels.


It was fertile ground for planting seeds of doubt. "There were countless nights in the building," said Maggio. "TI knew what the stakes were: If we got there first, there'd be a big payoff, so the pressure was enormous."


Staszewski toughed out the stress because he had a powerful motivation to participate: "I was working on my final dissertation for my PhD, and I wanted something novel." It so happened that Krenik was on his PhD committee. "I couldn't go back and say I was going to implement the chip in traditional CMOS," he said.


Tasked by Leipold with designing the frequency synthesizer and the transmit chain, Staszewski did produce some conventional proposals at the start. But Leipold kept throwing him out of his office, demanding fewer inductors.


"One inductor in a deep-submicron process can easily get to 50k gates," Leipold explained. "That's the same size as a full ARM processor with a little bit of memory."


"This was something that really shook us," said Staszewski, who was still busy brushing up on charge pumps and phase detectors. "It pushed us in this direction of saving as many analog components as possible and doing as much as possible in the digital domain."


To do so, the group decided to leverage the inherent advantages of deep-submicron processes, specifically their fast switching times and consistent capacitor matching. "We realized we could use a time-to-digital converter to go from a time to a digital value, bypassing the analog voltage and making everything much simpler," said Staszewski. Discussed at the 2004 International Solid-State Circuits Conference, the TDC concept developed into a readily controlled, all-digital-signal system that eliminated the need for a phase-locked loop to generate precise analog control voltage for the VCO.


The ISSCC paper described the technique as it applied to a 0.13-micron Bluetooth transceiver. Because of its relatively low performance requirements relative to GSM, Bluetooth became an excellent proving ground for the DRP concepts. The BRF6100 Bluetooth single-chip product was the first real-world implementation of DRP and was announced in June 2002. That chip was followed by the BRF6150 and the BRF6300.


Muhammad, for his part, had gotten roped into the project when Leipold walked into his office and asked, "If we were to sample RF, what would we do with it?"


"That got me figuring out the possibilities, as at deep-submicron processes, you can't get gain from transistors at RF frequencies," Muhammad recalled. He joined the team in 2000 as a "loaner" from the research group.


Right away, Muhammad figured out that there would be difficulties with the settling-time stats that would be required to get the necessary resolution, signal-to-noise ratios and sensitivity. "So, we figured the best thing was to move to the current domain," he said.


That leverages the fast switching time and capacitor consistency to perform switched-capacitor filtering, Muhammad said, "so that at the first point we are converting down to baseband, we're actually filtering out most of the energy that we are not interested in." The idea led to the multitap direct-sampling mixer (MTDSM), also presented at last year's ISSCC.

Functions combined
With the MTDSM, said Krenik, "you're getting heterodyning [down-conversion], gain and selectivity, all in a single operation. That's the power of Khurram's receiver concept."


Switched-capacitor filters themselves are not new; what's unusual here is the frequency of operation in the GHz range, made possible by the process node. "The architectures that we've developed simply would not work if we were back at 0.25 or 0.35 micron," said Krenik.


While circuit innovation was key, "there also had to be maniacal focus on getting to production, so every idea had to be proven," said Maggio. Validation thus was key, he said — "and there were some massive tool problems."


For example, "We had to trick the VHDL simulator," capable of resolution down to femtoseconds only, "into thinking it was operating at attosecond resolution," Staszewski said.


"We were making demands that had never been made before," noted Krenik. So, all were pleasantly surprised at the results of first-pass silicon.


The chip is still in characterization, but Krenik said it meets all critical GSM performance metrics. The next stage is to apply the techniques earlier in the next process node, at 65 nm. Krenik expects the first products based on that node to appear in 2007.


Another promise?