The rise of the single-chip radar
Image credit: rex features
High-frequency circuitry will bring radar to cars, medicine and rescue services.
Lars Reger, Automotive CTO at NXP Semiconductors, sees sensors and computing driving a transformation in car design, turning the vehicles into “robots with wheels”. One of the most important sensor technologies for this new generation of transport, in Reger’s view, is radar. Along with competitors like Infineon Technologies, NXP is taking advantage of improvements in silicon integration to make use of frequencies that not so long ago were practically out of reach of electronics designers. Reger calls the 77GHz to 81GHz range “the radar frequency of choice”, not least for the effect it has on the circuit design. “The higher the frequencies the smaller your antennas,” he says.
Whereas putting in an antenna for Bluetooth or Wi-Fi at 2.4GHz means trying to squeeze a wire into a shape that fits an unused corner of a PCB or enclosure - and is not often straightforward - the antenna for the millimetre-wave range of NXP’s favoured zone for radar can potentially fit onto the chip that generates and interprets the reflected signals.
Although the frequency range makes tiny antennas feasible and reduces the size of other necessary components, such as inductors, designing at such a high frequency in a mass-production semiconductor process such as those used for microprocessors presents challenges.
Although NXP had the design ready for a 60GHz short-range communications chip, Reger recalls the scepticism he encountered from his engineering team: “‘Lars, dream on’, they said. ‘It will work but not with the power output we need’.”
But the team told Reger just before Christmas 2013, less than a year after starting the project, they had transmitted the design to the fab with the responsibility to make the first samples. He says the plan is to combine microprocessors with multiple high-frequency front-ends.
“We see two flavours of radar system emerging. One is to have radar stamps distributed around the car. We can take four of those front-ends and integrate them onto one device. With four front-ends you can have beamforming radar. I call it high-brightness radar for our cars,” Reger claims.
New uses for radar
Automotive radar is just the beginning. As transmission frequencies move beyond 80GHz and into the so-called terahertz range that sits between the mm-wave and far infrared regions of the electromagnetic spectrum, researchers are finding many uses for radar that can create novel types of sensor. Because many materials that are opaque under visible light become transparent in the mm-wave and terahertz range, these frequencies are likely to prove important for non-destructive testing and also for non-contact sensing in medicine.
In January, Kyoto University researchers working with Japanese electronics manufacturer Panasonic unveiled work on a remote heart-rate sensor based on mm-wave radar. The idea is to develop ‘casual sensors’: devices that take measurements as people move around the home instead of having to be worn on the wrist or chest.
“Taking measurements with sensors on the body can be stressful and troublesome, because you have to stop what you’re doing,” says Panasonic researcher Hiroyuki Sakai. “What we tried to make was something that would offer people a way to monitor their body in a casual and relaxed environment.”
Toru Sato, professor of communications and computer engineering at Kyoto University, adds: “Heartbeats aren’t the only signals the radar catches. The body sends out all sorts of signals at once, including breathing and body movement.”
The ability to sense breathing using radar could see the use of mm-wave radars in rescue operations. By taking advantage of frequencies that can penetrate dry materials like concrete but which are reflected by those with high concentrations of water, radar can peer underneath the rubble of collapsed buildings to find survivors.
One important characteristic of the recent developments in mm-wave and terahertz sensors is the shift towards putting all the radio elements on-chip. This can lead to near-field imagers that use terahertz waves to identify materials and flaws in them. One novel application is an alternative to X-rays for identifying the layers in paintings, down to the brushstrokes used.
One reason why the mm-wave and terahertz regions have gone unexplored for so long, outside of specialist and often military applications, is simply the mismatch between low-cost technologies and capability. The antennas might be of the right size for integration but the amplifiers behind them need to operate at frequencies more than ten times higher than those used in cellular phones.
Gradual reductions in transistor size over the past 50 years have increased their maximum switching frequency into the mm-wave range. Usable terahertz designs do not even need the leading-edge processes in the sub-20nm range used for high-speed microprocessors: they can employ what are now considered mature technologies.
In 2012, researchers from Cornell University used a 65nm silicon process - one that foundry TSMC introduced ten years ago - to build a 260GHz transmitter. Later iterations improved the output power by moving to modified process that adds germanium, commonly used in RF power amplifiers, to boost the power output to a level suitable for near-field imaging sensors.
The performance of active devices like transistors and diodes has helped complete the implementation jigsaw. Sharad Kapur, president and co-founder of RF design tools supplier Integrand Software, points out that a number of process developments over the past couple of decades have made it easier to implement the passive components needed to complete high-frequency circuits. For example, to support Bluetooth and other short-range radios working in the gigahertz range, chipmakers made it possible for designers to include inductors using thick strips of metal wiring on the uppermost layers of integrated circuits.
The move upward in frequency means that constructing many of the passive elements on-chip, such as capacitors and inductors, becomes easier. “When it comes to inductor sizes, above 60GHz they turn into little stubs of metal,” Kapur says. “It’s also very difficult to build off-chip passives at very high frequencies. It’s simpler to build them on-chip.”
Similarly, the tiny wavelengths of these sensors and radars match antennas that can be laid out entirely on the die of a silicon chip. But there is one big downside to multi-gigahertz frequencies and on-chip integration: design becomes more troublesome. Every RF designer has to worry about the electromagnetic coupling between components and wiring that may create unwanted parasitic elements.
Ansys director of product management Larry Williams says: “At mm-wave frequencies, every piece of metallisation is a potential antenna, inductor, capacitor, or a little of each. Designers cannot avoid the physics and hence much of the design has potential to be more difficult.”
David Vye, director of technical marketing for National Instruments’ AWR design software group, says: “As frequency increases, parasitic series inductances and resistances as well as shunt capacitances also increase. The resulting impact on circuit performance includes lower gain and output power, inferior phase noise performance, lower power-added efficiency, reduced linearity, instability and more.”
Kapur notes that at such high frequencies the interaction between the circuit structures becomes much more complex. “They can couple to faraway structures,” he points out.
The realities of on-chip integration bring other problems. Silicon processes such as those based on CMOS structures make it possible to integrate more functions onto one device than those based on materials, such as indium phosphide, that offer higher electron mobility and so reach further into the terahertz spectrum. Indium phosphide devices have been demonstrated that double the transition frequency (fT) over that of CMOS. But silicon as a material has its downsides.
“Modern CMOS processes can provide sufficient fT for mm-wave applications, but there are numerous challenges,” Williams warns.
Vye points to the electrical properties of the silicon substrate itself, which reduces the quality factor (Q) of devices such as capacitors and inductors. This lower Q translates into more noise and higher power consumption. Kapur adds: “Capacitors can be the bottleneck now. They become low-Q and you have to start worrying about losses. They also start to work like inductors and you get resonances.”
Williams says the push for integration leads to layout approaches that make the job of design harder. He points to a seminal paper written in 2009 by UCLA professor Behzad Razavi that compared design techniques based on silicon and on III-V materials, which include indium phosphide and gallium arsenide. Faced with the need to keep die size down and leave room for logic functions and other analogue circuitry, the silicon designers tried to save space using more inductors that neatly fold into spiral shapes. Those working on III-V materials tend to favour approaches used for years in microwave design, employing long transmission lines rather than inductors. The difference in size can be dramatic, with differences in overall circuit area of as much as twenty-fold.
The differences in approach mean that, as frequency scales up, the reduction in area of devices such as inductors is not as large as might be expected. A spiral inductor for 60GHz operation turned out to be half the size of one operating at a tenth of that frequency. Core transistor speed had only scaled by around five times, which made the floorplanning and layout of the circuitry much more fiddly.
“Co-design is key,” argues Wiliams. He says the job of making these circuits work needs to include physics-based modelling alongside circuit design together with system-level analysis. “Consider a 77GHz radar module for example. At its heart there are CMOS transistors at the nanometre scale. Beside them are spiral inductors at the micrometre scale, and IC packaging and embedded antennas are at the millimetre scale. Assembling a system model that traverses this giant range of dimensions while including effects at each level is a big challenge for designers and design tools.”
Designers have the additional issue of pushing processes into a regime for which the manufacturers have not planned in advance. The models and process design kits (PDKs) they supply to designers for use in simulation are generally characterised for analogue and RF circuits used in mainstream design - which today is well below 10GHz. Vye says work is ongoing to develop PDKs with manufacturers that reflect the realities of mm-wave and terahertz design but that even with them in place, electromagnetic (EM) simulation tools are vital.
“EM tools are playing an increasingly important role in on-chip component characterisation as well as partial and entire IC design verification. The challenge foundries and their customers face is to have EM-simulation-ready component models,” Vye explains.
PDKs are important, he says, because the models need to reflect the details of how the chip is made, such as how the metal layers above and below an inductor are formed, for example. “A structure such as an inductor ground ring, which may be well-defined by a layout mask for processing, may not be suitable for EM analysis without modification,” Vye says.
Williams says that even with PDKs, electromagnetic simulation is essential. “Because the interconnect is a larger fraction of the layout in mm-wave circuits, the full modelling of spiral inductors with the associated interconnect using EM simulation is critical. There simply are no models in the PDK that account for the coupling among inductors and interconnects. Nor are there models for integrated antennas. Antenna gain and efficiency is hard to achieve but can be fully predicted by EM solvers. Clearly this is a great opportunity for simulation to fill in the gaps,” he says, pointing to the desire to integrate antennas on-chip and save cost.
There remains a tension between the needs of foundries to protect their trade secrets and still provide enough openness to make high-integration mm-wave devices possible.
There are a number of techniques that chip manufacturers use to improve yield, such as depositing areas of metal that are not connected to anything directly. This dummy fill helps ensure metal lines are not etched away accidentally by later processes and to maintain a flat surface for the layers deposited on top. The unconnected metal lines are still affected by EM fields, and develop floating voltages that interact with other components that are connected to power rails.
Kapur says: “The floating metal set up reverse eddy currents, which increases the complexity of the problem.”
The details of how the metal layers are deposited and etched, however, can provide competitors with useful information on the capabilities of the process. Foundries are keen to protect that intellectual property.
Williams picks up the same point. “Recently, some CMOS foundries have encrypted their layer stack-up to protect their process IP. Ansys is currently working with these foundries to enable leading semiconductor companies to use physics-based 3D modelling while simultaneously hiding the layer stack,” he says.
Although the design challenges present obstacles to adoption, the new territory of sensors and radar based on mm-wave and terahertz transmission is beginning to open up thanks to single-chip integration. As tool offerings expand, the design problem should become easier in the coming years. *