2019 IEEE IEDM event reveals latest node chips, chiplets, memories for AI, densest thin-film batteries, 400Gbits/s silicon photonics, quantum computing tools and much more.
The theme for this year’s 65th IEEE International Electron Devices Meeting (IEDM) was, “Innovative Devices for an Era of Connected Intelligence.” As in previous years, major semiconductor players including and international research organizations (e.g., imec, CEA-Leti, UC universities and others) presented the latest detailed technology for processors, memories, interfaces and power device devices. Additionally, the event included quantum computing advances, medical uses and other newer areas of application.
Here are 10 of the major semiconductor “reveals” at the show for 2019.
Leading Edge 5nm Chip with Super Dense Memory
Moore’s Law may be hitting the wall but it’s not dead yet. TSMC unveiled a complete 5nm technology platform that advanced silicon chip scaling (miniaturization) to the next process node. Reaching the 5nm node milestone was due in part to advances in lithography and improvements in process and packaging techniques.
TSMC researchers described a 5nm CMOS process optimized for both mobile and high-performance computing. It offered nearly twice the logic density and a 15% speed gain or 30% power reduction over the company’s 7nm process. The process optimization incorporated extensive use of EUV lithography to replace immersion lithography at key points in the manufacturing process.
TSMC’s 5nm platform also featured FinFETs and high-density SRAM cells. The SRAM could be optimized for low-power or high-performance applications, and the researchers say the high-density version was the highest-density SRAM ever reported. The researchers say high-volume production was targeted for 1H20.
Quantum computing
Great strides have been made in quantum computing. At the Semicon West/Electronic System Design (ESD) 2019 conference, IBM displayed it’s IBM Q Experience, a cloud-based quantum computer available for free to anyone with a web browser and an internet connection.
Creating a quantum computer has been an amazing technological achievement, but like any computer it needs software. Imec – the international Flemish R&D nanoelectronics organization – presented the first step toward developing a systematic approach to the design of quantum computing devices.
EDA chip design software such as TCAD is necessary to produce highly accurate models of semiconductor devices and their operation. To date, no analogous tools exist to model qubits, the basis of quantum computing, because the field is so new and complex. If these design tools did exist, the development of quantum computers could take place much more quickly.
The Imec team has taken a step to create such a software framework using multiphysics simulation methods to develop a comprehensive design methodology for qubits built in silicon. They modeled device electrostatics, stress, micro-magnetics, band structure and spin dynamics. Based on the results of these studies, they say that single-electron qubits in quantum dots can be induced and optimized in silicon MOSFETs with thin (<20nm) gate oxides. The researchers will discuss critical aspects of their methodology, the parameters they modeled, and next steps.
3D Chiplets
Intel presented a novel 3D heterogeneous integration process for chiplet creation. It is seen as an evolution of Moore’s Law, a way to keep the scaling, size and cost benefits continuing into the foreseeable future.
Chiplets are a type of advanced packaging which offers a different way to integrate multiple dies into a package or system. There are a number of ways to make chiplets, but all use a library of modular chips – like Lego building blocks. These module chips are assembled in a package that connects them using a die-to-die interconnect scheme.
There are many other approaches to combining chip dies, i.e., 2.5D dies that are stacked on top of an interposer. But the hope with a chiplet approach is that it’s a faster and less expensive way to assemble various types of third-party chips like processors, memory, interfaces and the like.
Here are the details: Intel believes that heterogeneous 3D integration will drive scaling. CMOS technology requires both NMOS and PMOS devices. Intel researchers used 3D sequential stacking architecture to combine these different devices. They first built Si FinFET NMOS transistors on a silicon wafer. On a separate Si wafer they fabricated a single-crystalline Ge film for use as a buffer layer. They flipped the second wafer, bonded it to the first, annealed them both to produce a void-free interface, cleaved the second wafer away except for the Ge layer, and then built gate-all-around (GAA) Ge-channel PMOS devices on top of it. The researchers say these results show that heterogeneous 3D integration is promising for CMOS logic in highly scaled technology nodes.
This images hows a schematic and a cross-section of a fully processed 3D CMOS transistor structure achieved by this process; in the middle is a thickness contour map of the Ge transfer layer, showing good uniformity; and at right is a 3D cross-sectional view of the completed 3D CMOS chip showing Ge-channel GAA transistors on top of Si FinFET NMOS transistors.
AI That Does’t Forget
Embedded STT-MRAM and other non-volatile memories (NVMs) are getting a lot of attention lately. NVMs devices retain their memory even after the power is removed. Embedded SST-NRAM is one NVM that shows particular promise in the embedded memory space for cache memory in IoT and AI applications.
At IEDM 2019, TSMC described a versatile 22nm STT-MRAM technology for AI while Intel talked about STT-MRAMs for use in L4 cache applications.
In STT-RAM writing, an electric current is polarized by aligning the spin direction of the electrons flowing through a magnetic tunnel junction (MTJ) element. Data writing is performed by using the spin-polarized current to change the magnetic orientation of the information storage layer in the MTJ element. Intel improved the process and stack for L4 cache applications. STT-MRAM technology for L4 cache requires tighter bitcell pitches, which translate into smaller MTJ sizes and reduced available write current.
Organ Forceps With a Special Touch
Our internal organs are slippery because they’re covered with blood and other body fluids, so grasping and pulling them with forceps can be challenging. Although contact-force sensors have been placed on the tips of forceps used in diagnostic laparoscopic and robotic surgeries, there currently is no way to know if they are slipping, other than visually via a monitor, which has limited usefulness. A Kagawa University team described a highly sensitive slip-sensing imager (sub-mm resolution) and novel algorithm that can, in effect, give forceps a sense of touch. The idea is to use the device to visualize the spatial distribution of the grasping force across the organ’s surface. The center of that distributed load is calculated, and as the forceps are moved the algorithm relates any corresponding movements of the load center to slippage. Built on an SOI wafer, the device’s force-sensor pixels consist of a 20µm–thick piezoelectric silicon diaphragm (400µm diameter) with a center contact, and with a force detection circuit integrated on the diaphragm. The diaphragm acts as a strain gauge as it flexes due to varying grasping force.
Impedance Sensor for Fingerprint Imaging
Researchers led by Cornell discussed the monolithic integration of a piezoelectric aluminum nitride (AlN) resonator into a CMOS-controlled, GHz ultrasonic impedance sensor/imager. The device measures changes in surface properties such as surface oxidation, materials, liquid viscosity and others, and is meant for use in wearable, IoT and smartphone systems to detect fingerprints with high resolution, determine tissue states, and for other applications. This is the first time monolithic fabrication – all in one chip or die – has been successfully demonstrated, and it led to small, power-efficient GHz sensing arrays with improved performance vs. the standard two-chip heterogeneous integration approach, thanks to less parasitic coupling and a higher signal-to-noise ratio.
Thin-Film Battery Goes High-Density
The miniaturization of power sources hasn’t kept pace with the miniaturization of electronics. Although integrated electrochemical capacitors offer high power density, high frequency response and novel form factors, their low energy densities are of limited value for MEMS and autonomous device applications that require long periods between charging. CEA-Leti researchers discussed a thin-film battery (TFB) with the highest areal energy density yet reported (890 µAh/cm-2) and high-power density (450 µAh/cm-2). Built on silicon wafers using UV photolithography and etching for the successive deposition and patterning of each layer, the thin-film battery integrates a 20µm-thick LiCoO2 cathode in a Li-free anode configuration. It showed good cycling behavior over 100 cycles, and the fact it was built using a wafer-level process opens up the possibility to tightly integrate this battery technology with future electronic devices.
Physically Unclonable Function (PUF) for Mobile and Smart Devices
The spread of networked mobile devices and smart gadgets in the IoT landscape has created an urgent need to protect them with lightweight and low-power cryptographic solutions. A physically unclonable function (PUF) is a hardware-intrinsic security primitive, or basic programming element. UC Santa Barbara researchers discussed an ultra-low-power PUF that operates on the varying electrical resistances and current leakages that arised from intrinsic process variations in ReRAM crossbar arrays. The team built 4K-ReRAM passive crossbar circuit arrays fabricated with a CMOS-compatible process suitable for back-end-of-the-line (BEOL) integration. The arrays allow for an extremely large number of challenge-response pairs (a common cryptographic protocol), as well as 4x better density vs. other ReRAM architectures plus a ~100x improvement in power efficiency and more robust security metrics.
Silicon photonics
Very fast speed data races around within data centers via optical fiber, using silicon photonic (light-based) interfaces that operate at 100 Gb/s. But cloud data center traffic is growing at nearly 30% per year and there soon will be a need to increase the data rates. A STMicroelectronics-led team described a new silicon photonics technology platform built on 300mm Silicon-on-Insulator (SOI) wafers, yielding devices that operate at 400Gbits/s (each device has 4 channels, each of which operates at 100Gbits/s, for a total of 400Gbits/s).
Optical coupling and polarization management are key requirements, and their devices incorporate a 60 GHz high-speed photodiode and a high-speed phase modulator. They also built devices with a supplementary SiN waveguide layer for higher coupling efficiency, to meet evolving data-transmission requirements. The researchers say the photonics platform has the potential to meet the requirements of applications other than data centers, too, such as automotive.
The image is a photo of the chip-on-board assembly of an analog front-end (AFE) function implemented in a 400G-DR4 optical transceiver using the technology, and (2b) are PAM4 signal eye diagrams at 106 Gbits/s per channel, used to measure high-speed signal quality.
5G and beyond
One of the challenges for chip makers is how to integrate III-V materials with silicon to make ultra-fast devices for 5G and other uses, which are compatible with conventional CMOS technology. In addition to silicon, III-V compound semiconductors are obtained by combining group III elements (essentially Al, Ga, In) with group V elements (essentially N, P , As, Sb). This gives us 12 possible combinations; the most important ones are probably GaAs, InP GaP and GaN.
IOT and 5G applications typically use sensors that transmit wireless data to anedge or cloud network. This requires a combination of RF capabilities with a small form factor and low operating power. A promising approach to achieve this combination is to create single chips that combine the capabilities of silicon CMOS with those of III-V devices, such as gallium nitride (GaN) and indium gallium arsenide (InGaAs). The unique properties of III-V compounds make then well suited for optoelectronics (LEDs) and communications (5G).
At IEDM, Intel talked described how low-leakage, high-k dielectric enhancement mode GaN NMOS and Si PMOS transistors were built monolithically on a 300mm Si substrate. The goal was to combine GaN’s high-frequency/-temperature/-power attributes with silicon CMOS circuitry’s digital signal processing, logic, memory and analog capabilities, to create compact devices for next-generation solutions for power delivery, RF and system-on-chip (SoC) applications. The researchers say both device types demonstrated excellent performance across a range of electrical specifications.
III-V materials offer higher electron mobilities than silicon, and HBTs made from them are very fast transistors often used for RF and other high-frequency applications. A key goal is to build them on 300mm silicon wafers instead of other substrates, to take advantage of silicon’s lower manufacturing costs. A team led by imec described how they used a unique nano-ridge engineering technique to build GaAs/InGaP HBTs on a 300mm silicon substrate.
RELATED ARTICLES:
John Blyler is a Design News senior editor, covering the electronics and advanced manufacturing spaces. With a BS in Engineering Physics and an MS in Electrical Engineering, he has years of hardware-software-network systems experience as an editor and engineer within the advanced manufacturing, IoT and semiconductor industries. John has co-authored books related to system engineering and electronics for IEEE, Wiley, and Elsevier