by Greg Turetzky
I started my relationship with GNSS and Moore’s Law in 1985, writing software for GPS tracking loops on the Advanced Range Instrumentation Aircraft program at the Applied Physics Laboratory of Johns Hopkins University for the U.S. Air Force. The project’s purpose was to navigate a large jet to accurately fly a pattern to drop buoys into the ocean. That receiver had seven circuit boards (six trackers and one navigator) mounted on a VME backplane in a 19-inch rack mount in the back of a C-130, and was about the size and weight of suitcase.
In 1988, I helped design and build a single-board Swordfish receiver at Stanford Telecom that went into a two-man portable pseudolite for Trident missile testing. This was considerably smaller and lighter: about the size and weight of a desktop computer. Moore’s law — which, by the way, states that the number of transistors that can be placed inexpensively on an integrated circuit doubles approximately every two years — helped mostly by allowing much better CPUs and memories so we could put it all on a single board. I actually carried this beast off a landing ship tank (LST) onto a small island in the South Pacific called Kwajalein.
With Moore’s law in full swing in 1990, I moved to the commercial sector at Trimble Navigation and worked on the NavTrac, a lunchbox-sized complete GPS receiver for marine navigation, and then onward to timing receivers and eventually credit-card-sized modules. It became clear that Moore’s Law was a great friend of GNSS and was going to enable a whole new slew of applications by moving from the board level to the chip level.
I went to SiRF Technology, Inc., very soon after it was founded in 1995, to help develop the first commercially successful GPS chipset, the SiRFstarI (see photo).
SiRFstarI-based module, both sides, with representative AA battery to scale.
You can see that this module still had separate chips for the CPU, flash, SRAM, GPS correlator chip, the GPS RF ASIC, and a lot of other components.
Last year, we introduced the SiRFStarIV architecture and the GSP4e chip. The module made from this chip has the same basic functionality (RF in, position out) but at a much higher performance level in terms of sensitivity, time to first fix, accuracy, and much lower power consumption. The photo at right shows a 4e module. Also note how few external components are required.
To really understand the impact of Moore’s law on GNSS today, we have to break down the impact on the various parts of the receiver. The measurement of each section (area, power, or bytes) was then normalized to a starting point of 100 in 1995. The time span of 14 years is about seven Moore’s law doublings (every 2 years), producing an expected decrease of 1/128. We can see that the power and digital silicon area have tracked very well over that time period. However, it is also apparent that RF has not even come down by half in that time frame (although it has swallowed a lot of external components as seen in the pictures) — and the code size (ROM + RAM) has grown by 2.5 times.
This has turned Moore’s law into a bit of a foe in the current timeframe, as the costs associated with silicon products are clearly known to customers (die size is easy to measure) and has driven the prices for GPS receiver downwards accordingly. However, as one can see, more and more software is needed to enable the new features and functions, and with dropping prices due to decreased silicon size, it becomes harder and harder to pay to feed all the hungry engineers here at CSR. This is the crossroad at which our segment of the industry has arrived: how do we continue to add innovation and still make a profit selling silicon when Moore’s Law is not helping anymore? I am not sure I know the answer yet, but we have a lot of good ideas that we are working on.
Most of these ideas come from expanding the notion of location determination to extend beyond using just GPS and its currently available augmentations. Adding support for other GNSS constellations requires more hardware; the amount is highly dependent on which constellation(s) we are talking about. GLONASS, because of its different frequency, requires more RF silicon, requiring more total area because the existing area is not shrinking as fast. Galileo and COMPASS will require more digital area for their complex coding schemes, but these can be more easily handled with shrinking process geometry. All will require significant software effort to bring in new acquisition schemes, tracking loops, and navigation algorithms.
But location determination will not be a GNSS-only problem for much longer. Hybrid navigation using other signals of opportunity and MEMS sensors will play a large role in expanding the ability to provide accurate location to consumers wherever they go. The integration of these technologies into a coherent location determination system is a large software effort, and one that CSR has been working on for years in automotive applications.
Clearly, the need for accurate location continues to grow in consumer devices. At CSR we feel we are in the best position to deliver that, with or without help from Moore’s law.