philadelphia — before we see 5g networks and handsets, semiconductor fabs will need to change the way that they process wafers, test engineers will have to figure out how to test, and handset designers will have to figure out how to track steered beams as people move. on top of these and other problems, wireless products will have to sell for prices close to current levels. these were the takeaways from the 5g summit at the 2018 international microwave symposium (ims).
while some 5g ics are starting to appear, it's still unclear which processes will win for the fabrication of power amplifiers (pas) and phased-array antennas. regarding pas, participants in the panel "mmwave radios in smartphones: what they will look like in 2, 5, 10 years" discussed the so-called "iv" processes (silicon cmos and sige), and the "iii-v" processes (inp, gaas). iv are elements in column 14 of the periodic table (si), while iii-v are in columns 13 and 15.
session moderator dylan williams of nist noted that inp outperforms cmos pas at high frequencies (mmwave, in this case), while cmos wins at frequencies below 6 ghz. but harish krishnaswamy of columbia university noted that circuits made with an iii-v process are more efficient than those made from a cmos process. furthermore, "iii-v processes fail today in high-speed switching," added deveraux palmer of lockheed martin. that limits their use.
"does efficiency matter?" asked williams. the answer came in that a handset needs to last 1½ days on a charge, which lets you get to the next day when you forget to charge the phone battery overnight. you can still charge it the next morning.
ims 5g summit panelists (l–r): tim larocca, northrop grumman; joy laskar, maja systems; gary st. onge, anokiwave; harish krishnaswamy, columbia university; dev palmer, lockheed martin; farshid aryanfar, straighpath communications; walid ali-ahmad, uscd; amarpal khanna, national instruments (moderator); dylan williams, nist (moderator). photo by martin rowe.
"frequencies above 6 ghz will require breakthroughs," said macom's anthony fischetti in a later presentation. "the iii-v processes are different than cmos and power is too much for gaas at frequencies below 6 ghz." fischetti explained how his company, macom, is coping with the different processes. for example, macom is partnering with stmicroelectronics to make rf devices using a gan-on-silicon process. while such a process is possible, it's impractical at the production quantities that will be needed. the equipment needed either isn't available or is extremely expensive. he notes that macom currently manufactures some 50,000 cmos wafers a week at a fab running 24/7. with today's equipment, the company would need a month to make that many gan (iii-v) wafers. "iii-v fabs have to change to ramp up to the scales to today's cmos processes."
fischetti also noted that for a iii-v process to be economically feasible, it can't have reworked wafers. the quality has to be built into the process. plus, photolithography has to be used to shoot the images of the layers onto the wafer. electron-beam lithography is too slow. another issue with iii-v processes is that there can't be any gold in the clean room. employees must remove all gold watches and jewelry.
in addition to the process issues that 5g will bring, there are test issues. in the panel session, loy laskar of maja systems said that 80% to 90% of a bill of materials (bom) can be taken up by ic assembly and test.
while testing ics and systems made from exotic processes, mmwave frequencies, phased-array antennas can be done for military applications in which volumes are low, such testing at the high volumes needed for consumer devices is quite a challenge. charles schroeder of national instruments and roger nichols of keysight technologies highlighted some of the testing challenges, the most prominent stemming from the fact that testing may need to be done using over-the-air (ota) methods. probing mmwave systems with highly integrated devices (pas and phased-array antennas) may be impossible. ota testing will surely impact production test times. plus, the test equipment needs to handle the load.
schroeder noted that the math needed to process the wide-bandwidth signals that mmwave frequencies bring is enormous and that takes time. at this point, test engineers don't know if they will need pc-class processors, fpgas, or gpus to handle the signal processing. that will take some rethinking of the way that wireless signals are processed today.
other issues arise from high bandwidths. because bandwidths are so wide — perhaps 100 mhz — impedances of transmission paths can vary. test systems will have to be aware of that and compensate accordingly.
nichols took test issues a step further, citing issues of near-field-and far-field measurements. "ota testing in the far field could be construed as more difficult than in the near field. there is, however, a tradeoff. in the far field, the electromagnetic fields are better behaved. for example, this is the point at which we are close to the classic definitions of perpendicular e field, h field, and poynting vectors. what makes testing harder in the far-field is the signal loss and the size of a chamber. in the near-field, the challenge is getting an accurate sense of the antenna behavior and the relationship between phase and amplitude of the signals. furthermore, mmwaves have short wavelengths and because the near-field to far-field transition is proportional to the inverse of wavelength, the smaller the wavelength, the more distant is this nf/ff transition."
nichols noted that "people are random; they do random things like moving their handsets." such movement hasn't been an issue because antennas have been designed to be omnidirectional. to reduce power consumption in 5g, beam-steering with phased arrays will likely become the norm. that will force testing to be performed at different orientations, and test systems will have to verify that the handset will constantly track the handset's orientation and adjust the beam accordingly. on top of it all, measurement uncertainties must come down, said nichols. "you don't know how well something works until you can put numbers to it."