ISSCC Session 6 Medical and Sensor Technologies
February 18, 2013, ISSCC, San Francisco—Session 6 demonstrated emerging medical and sensor technologies. Contributors included KAIST in Korea, Masdar Institute of Science and Technology in UAE, National Chiao Tung Universtiy and China Medical University and ASE in Taiwan, University of Tokyo in Japan, EDA-LITEN in France, Eindhoven University of Technology in The Netherlands, ST Microelectroincs in Italy, Delft Universty in The Netherlands, Jet Propulsion Lab in Pasadena, and Stanford
Session 6.3 C.W. Chang from National Chiao Tung University presented a TSV-based neural sensor. The various challenges for neural sensors are probe density, probe reliability, size, and many others. For long-term monitoring, probes are implanted into the brain, and the monitoring system includes electrodes, interconnections and circuits.
The neural probes or electrodes are usually developed in a different technology process than the CMOS circuits. The long interconnections induce noise and increase packaging size as well as cause signal attenuation. As a result, new processes need to address the integration of the circuits, interconnect, and probes into a single part.
One way to achieve smaller packaging is to design a stacked chip or chip on board structure that is bonded to the interconnect and then to the probes. This work integrated the probes and circuits into a single 2-sided array that uses TSVs to achieve a reduction in path length leading to shorter path lengths, lower noise, increased signal capture, and reduced area.
This chip has 16 channels with 480 probes in a single two-sided device with TSVs. The interconnect is 42 TSV per channel. The fab flow was a standard 0.18 micron CMOS process with back grind for 2 micron TSVs. The backside material was retained for the probe structures. Ion etch formed the probes and exposed the TSVs, and the backside was metalized. The resulting TSVs are 200 microns long and 30 microns in diameter. Probes are tapered structures 150 microns long with a base diameter of 25 microns.
The circuit was designed for easier front-back alignment and had 3 x 14 TSV per channel for contacts to the probes. The probes were arrayed in a repeating 5×6 pattern. The probes were characterized and the TSV leakage to adjacent TSVs was found to be within desired limits.
Session 6.5 Sahel Abding from Eindhoven Uiversity of Technology described a 4bit A-D converter in a printed organic complementary process. The challenges in printing transistors on a substrate are limited process control and many low-temperature processes. Printed organic thin film transistors are moving to find many applications where mechanical flexibility and large area are needed as in displays, matrices of sensors, and RFIDs with sensors.
The basic process uses complementary, organic-metallic, thin-film transistors that are mostly good for digital circuits. The devices are deposited or printed on a substrate and get a gold sputter as a base. This layer is laser ablated and a self-assembly monolayer is deposited. An oxygen plasma cleans the surface and a silver top gate is applied. Resistors in this process are about 35k-ohms per square.
As a result of the active and passive device characteristics, the a-d converter design uses a 4 bit transmission-gate counter driving a R-2R ladder. The output is fed to a comparator. When the comparator switches, the count is output. The comparator design is optimized for low mismatch by combining a current mirror and an inverter.
Session 6.8 Max Shulaker from Stanford presented a carbon nanotube (CN) FET capacitive sensor. This is the first subsystem built entirely from CN FETs and required imperfection-immune design and robust CNFET processing. The CNFET have an average diameter of 1.2nm with minimum lithography feature size determining half pitch. This sub-assembly used 4nm pitch.
CNFETs have good design characteristics, such as high on-current, low power-delay product, and 400mV operation. The previous works found many problems with mis-positioning of the carbon nano-tubes (CNTs) and a high population of metallic or always highly conductive devices. To overcome the main issues of CNFETs, they analyzed the growth of the CNTs and developed a defect immune design.
The CNTs are deposited on a quartz substrate and transferred to a SiO2 working surface. They etch the extended gate contacts and the CNTs. Then they customized a number of the units in a standard library to make the first CNT IC. Previous work only made single devices or stand-alone logic devices.
To address the metallic CNTs, they developed a technique to remove 99.99 percent of all the metallic devices and only about 4 percent of the semiconducting devices. This selective removal process uses a chip-scale electrical breakdown. They apply a gate voltage to the array of devices on the substrate to turn off all the normal devices. Then they apply a high voltage, about 10.5 volts, to the array, which causes the metallic devices to overheat and fuse open.
This method provides a high yield of working CNFETs and associated small scale test circuits like a frequency modulator block and a digital PLL. The other part of the effort was to design CNT immune circuits. First they grow the CNTs which will always leave some randomness. Then they add extended gate and contacts across a number of CNTs and etch the extended gates and CNTs, leaving only those contacts and CNTs that are needed for the circuit. Finally they dope the P and N regions to get the working circuits.
As a result, they have a no per-unit customization and a immune CNT library of circuit elements. Circuit robustness was verified by building arrays of 400 inverters. On average only 19 failed, mostly for non CNT issues. Continuing to more detailed designs, they built a capacitive sensor interface.
The sensor interface uses a sensor controlled oscillator and a digital PLL to sense a differential capacitance. A reference channel has a 1nf capacitor and the same ring oscillator as the test channel. When a capacitive load appears on the sensor input, it changes the frequency of the oscillator. This frequency change is detected with a 1 bit digitizer and output. For all of the working circuits, there is a distinct and significant difference between the loaded and unloaded oscillators.
The capacitive sensor array was coupled to a mechanical hand that shook when touched. Not a very sophisticated demonstration, but significant in its ability to detect a capacitance change with the hand was touched.
Future efforts will focus on increasing CNT density, reducing CNT contact resistance, and getting better control of CNT doping. The CNT variations offer a unique opportunity for improvements and there may be some correlation of performance by layout.
