CMOS Challenges SiGe Radar Chips
Could imec's CMOS car radar chip prompt industry to move to pure silicon, asks Compound Semiconductor.
Implemented in 28nm CMOS and with a supply voltage of 0.9V, Imec’s continuous wave radar transmitter only consumes 121mW and is fully compliant with the spectral mask imposed by ETSI. [imec]
Just last month, imec unveiled what it described as the world's first 79 GHz radar transmitter implemented in 28nm CMOS and designed for automotive radar systems. While today's high-end vehicles may feature a two- or three-chip single SiGe radar system - typically used in adaptive cruise control - industry pundits predict strong market growth for such systems. And imec wants to be a part of it.
"High-end series vehicles such as BMW and Mercedes feature a single radar system but in the future we will see multiple radars in a car," says imec researcher, Wim Van Thillo. "We could see four radars in the front bumper, four in the rear bumper and this is going to be in all cars, not just high-end."
But what sets imec's chip apart from the millimetre-wave sensors developed by other radar chip manufacturers, Infineon and Freescale, is the fact it is fabricated in plain and simple CMOS. Could this tempt radar system suppliers away from SiGe?
The first commercial radar systems of the late 1990s were based on GaAs chips. But then Infineon jumped ship and started developing systems based on bipolar process SiGe chips.
These SiGe chips not only met the high speeds demanded by 77 GHz automotive radar, but enabled smaller and more cost-effective radar systems than had been possible with GaAs-based versions. Freescale has since delivered 77 GHz SiGe chipsets - fabricated in 0.18 micron BICMOS - for radar collision warning systems, and automotive radar developers have been keen to switch from GaAs to SiGe. But now CMOS is catching up.
"SiGe transistors are much faster than CMOS transistors at the same node, but CMOS is now possible because of more advanced [fabrication] processes," explains Van Thillo. "We're down to 28nm for CMOS whereas the most advanced SiGe nodes are 130nm. We've been focusing on implementing analogue functionality on plain digital CMOS and we believe we are the first in the world to present such an advanced chip."
Clearly imec's motivation for straight CMOS is integration. With researchers having demonstrated they can implement analogue transmitter functionality - analogue and mm-Wave - on CMOS, the race is now on to develop a single chip solution.
Admitting he and colleagues are 'not there yet', Van Thillo says: "We've also now taped the receiver to the transceiver... We're waiting on measurements and by the end of this year will have proof that it can be done in 28nm CMOS."
With full transmit-receive functionality on CMOS - for a single antenna - in hand, Van Thillo then hopes to scale up to multiple antennas with higher resolution by 2015. "And once we have that we need to integrate analogue to digital converters and so on, to demonstrate this can all be integrated onto a single chip."
Clearly a single chip CMOS solution brings the promise of vast cost reductions at larger manufacturing volumes, but is CMOS actually going to be that cheap? The NRE costs of a 28nm node process are very high - a single mask set can cost up to €2 million, but Van Thillo isn't fazed.
"We have several reasons to believe CMOS will be a good choice," he says. "[Process] costs are coming down rapidly and we are targeting products to hit the market in two to three years, depending on partners. And, of course, automotive radar is growing fast."
But will imec's 79 GHz radar chipset rival SiGe 77 GHz versions? For today's automotive radar developers, two key frequency bands exist; 77 GHz and 79 GHz. The former is well-established and used for the long range radar in automatic cruise control systems in high-end cars. Radar range reaches 200m, with a resolution down to 30cm and narrow field of view.
More recently, the 79 GHz band has been allocated in Europe, although worldwide harmonisation is expected. This band suits shorter-range, high-resolution radar systems, to be designed for relatively new functions that demand a smaller depth resolution and wider field of view, such as stop-and-go, pre-crash alarm and blind-spot detection.
Existing 77 GHz SiGe solutions offer both the longer range and short-range functionalities, albeit across several chips, so why bother with the 79 GHz radar? As Van Thillo says: "The 79 GHz band offers the 4 GHz bandwidth that is required to bring the range resolution down to a couple of centimeters."
"For future systems, manufacturers will need shorter range radar with finer angular resolution which translates into more antennas," he adds. "So to scale up, systems will need more chip area, and in larger volumes CMOS is very attractive."
Clearly the SiGe versus CMOS debate has some way to go. As Van Thillo highlights CMOS could be used for non-critical radar applications until automotive manufacturers gain enough confidence to switch from SiGe. But then manufacturers could always stick with SiGe chips for longer-range radar and rely on CMOS for short-range applications.
However, he is clear that there is no future for the original GaAs radar chip.
"The end of GaAs in automotive radar has already happened," he says. "I see no reason to do things in GaAs that can be done in SiGe, there just isn't any competitive advantage here."