Maximus R&D
  • Quick Navigation
  • Home
    • About Pieter Hooijmans
    • About Maximus-R&D
    • Experience >
      • Radar Technology
      • Optical communications
      • Tuners and RF Modules
      • RF IC's
      • Communication Systems
      • Audio and Analogue
      • IC Technology
      • Packaging
    • R&D Processes
    • Services >
      • Client Projects
    • Contact
  • Technology History
    • Piet Hooijmans 1918 - 2006
    • Piet's Home-built Television pt1
    • Piet's Home-built Television pt2
    • EQ40 and EQ80
    • TV Tuner history pt1
    • TV Tuner history pt2
    • TV Tuner history pt3
    • Philips TV remote controls, 1955-1985
    • TV Tuner history pt4
    • TV Tuner history pt5
    • TV Tuner history pt6
    • Digital circuit blocks
    • TRANSDECO
  • Oil Painting
  • Vliegen met Pieter

IC Technology

Technology definition & development

I am not a process technology expert by profession, so for discussions about HOW to make an IC-technology I have to refer to the physics colleagues. However, over the years I've been heavily involved and responsible for technology analysis and definition. Answering questions like:
  • do we need a next generation technology?
  • what will be the main drivers for that next Generation? E.g. size or cost reduction, power consumption, performance improvement, feature integration (embedded memories, passives, sensors, ...).
  • make or buy (from a foundry)?
  • big steps or multiple small steps?
  • and of course associated investments (project cost and equipment), resulting wafer cost and the resulting business case for the technology development.
Having answered these questions always results in a business-driven technology roadmap, one of the most critical elements of every technology-centric company. And one of the most neglected elements by many businesses!
Below a summary of the many Si-technologies I've worked with.

RF Technologies

Dedicated RF performance in combination with silicon integration has always been driven by bipolar technology.
  • HS4 was the first breakthrough, a bipolar-only technology pioneered in the Albuquerque fab of Philips Semiconductors with a then (1991) staggering 14GHz fT. I was involved in optimizing the broadband performance.
HS4 developed into the first BiCMOS node, Qubic1 based on 50um CMOS and the same bipolar transistor. Via Qubic2 (18GHz) this developed into 0.35um Qubic3 in 1998, with an fT around 32GHz. This technology was the work horse for the first generations mobile transceivers. When I came back to the NatLab the next generation was in definition:
  • Qubic4, based on 0.25um CMOS and a double hetero-junction bipolar of 48GHz. My Research group was heavily involved in defining and piloting this technology.
  • Qubic4G was the first SiGe based technology, although using what some called a "homeopathic" addition of Germanium. Ft improvement was indeed modest at 60GHz. My Research group was the first and one of the few actually designing in this technology.
In my new role as RF Program Manager the first thing I did was driving the technology business case for a real SiGe microwave technology:
  • Qubic4X, a real microwave technology, supporting new applications above 5GHz. Based on a SiGe:C bipolar with an fT/fmax of 137/180GHz.
Based on the successes with Qubic4X the knowledge was used for two further optimizations:
  • Qubic4+, an upgrade of Qubic4 with especially a number of improved passives, thick top metal, DTI and a high-density but robust MIM-capacitor.
  • Qubic4Xi. Because much business was based on LNA's, the technology was further fine-tuned for lowest noise figure. ft/fmax 180/200GHz.
Foto
Typical Qubic SiGe bipolar transistor cross section, showing the lateral structure as opposed to the typical vertical MOS build-up.
Foto
The Qubic4Xi ft-curve in red, showing the effective 200GHz. For comparison in green the curve for 65nm RFCMOS. Although the intrinsic fT of deep sub-micron CMOS is high, the large interconnect parasitics reduce the effective fT to values half those of bipolar.
Foto
A typical Qubic4X microwave IC, with clearly visible high-quality inductors. This is a satellite down-converter operating at 10-12GHz.
Foto
Typical Gen7 silicon LDMOS power transistor cross-section.
Of course there are more RF-technologies than BiCMOS. Especially for RF power applications dedicated technologies are required. NXP (now Ampleon) has been one of the two leading players in LDMOS Si-based RF power transistors, as used in all cellular base stations. Over successive generations efficiency and (passive) integration level were driven to increasingly higher levels.
Where LDMOS is no longer efficient (i.e. above a few GHz or at extremely high power densities) III-V technologies come into play. Historically this was GaAs, which we used e.g. for mobile phone RF power amplifiers, the last years a lot of effort went into GaN. On this last one the RF efficiency advantage is undeniable, but the excessive cost structure is a major issue. Time will learn whether GaN will really become as successful as predicted.
RF-CMOS is a quite different technology segment. Here it is not the question how the technology can be optimized (it is simply given as a certain node of baseline CMOS), but how RF-design can still be done given the mediocre technology. Although the intrinsic fT of a CMOS transistor can be very high, the parasitics due to low-ohmic substrates and highly dense interconnect are typically high, reducing the effective performance considerably. The seriously higher design effort, the much higher mask set cost  make that RFCMOS solutions are effectively only viable in the specific case of single-chip SoC integration of an RF-transceiver with a much larger digital baseband/controller. Where the resulting medium RF performance must then be accepted as one of the consequences.
RFCMOS yes or no remains one of the most heated debates with respect to technology selection, and will probably remain so for a while. I went through these discussions for the 180, 140, 90, 65 and 40nm nodes, with grey hairs as a result.
Foto
A typical RFCMOS IC, with upper left the RF section, upper right the filter, PLL and power control, and the lower half the purely digital baseband. This one was in 40nm.

Power technologies

Like RF, power is a very broad technology area, with many different optimizations for the many power segments. Roughly speaking there are two main segments of power technology:
1. High-voltage/high-power discrete technologies
  • EZHV, an SOI 700V and later upgraded to 900V technology, enabling to connect circuits directly to (doubly rectified) mains
  • Trench-MOS; multiple generations of medium voltage power switching mosfets for predominantly automotive power applications
  • GaN-on-Si; the Trench-MOS successor. Using GaN much higher efficiencies can be obtained, where the challenge was to get a cost-effective technology into a Si fab.
2. Medium power integration technology, where power transistors at different higher voltage levels (typically 5, 10,20, 50, 100V) are integrated into (older) CMOS baseline nodes.
  • C50-PMU, our first power integration technology, as the name says used for PMU- and power control related applications
  • C14P, the successor, with a much denser digital node for integrating more control options
  • ABCD3 and ABCD9, roughly spoken versions of the above on SOI for higher reliability required in automotive applications.
Foto
An EZHV 800V die.
Foto
A GaN-on-Si 600V power transistor.

Foto
A silicon platform structure within a defined CMOS node, making application-specific optimizations by adding or deleting process modules.
Foto
CMOS mask build-up model to simulate wafer and mask cost for each of the CMOS nodes.

CMOS

In CMOS there are (for companies like NXP with internal wafer fabs) two types of technology strategy discussions around CMOS:
1. Internal platform extension
Although all companies will at one moment stop internal development of the next CMOS node, having the older internal CMOS platforms has a major advantage with respect to cost and supply. Within NXP the last and most cost effective platform was 140nm, ideally suited for the high-performance mixed-signal products we were making. As the main business using this platform I have driven to broaden it as much as possible, adding application-specific features:
  • NVM, adding a flash non-volatile memory
  • Power adding 5 and 20V power transistors
  • RF, with high-ohmic substrate, deep N-wells, a MIM-cap and thick top metal for improved RF-performance
  • Sensor, adding sensors (e.g. humidity and pressure) on top of baseline C14
  • Automotive, the SOI-version of Power with additional 100V transistors
This broad family structure made 140nm a very successful platform, used by almost all NXP business and responsible for loading our biggest fab (plus external 2nd sourcing).
2. Foundry CMOS sourcing
For baseline CMOS, offered by the foundries (TSMC, Global Foundries, UMC and more) the challenge is to determine (on company level) which nodes are the most efficient to use. This requires structural cost analysis, strategic prediction of what will be become the main nodes (both at market level and within the company) and analysis of the performance and features. Within NXP we went through this process for 110, 90, 65/55, 45/40 and 32/28 nm, within a Maximus consultancy project also down to 16 and 10nm.

Passive Integration and sensors

Because of the need for technology differentiation I have always been heavily involved in passive integration, essentially integrating high-performance passive functions on silicon. This could be both stand-alone passive technology, as well as integration of passive functions on top of a silicon platform. Some examples:
  • PASSI, a technology to integrate high quality RF resistors and inductors used in mobile PA modules
  • PICS, a technology to integrate high density (RF) capacitors on Si, and used as carrier for active IC's in an effort to reduce PCB footprint (business sold to IPDEA in Caen (France)
  • BAW filters
  • RF-MEMS switches for mobile phone band and antenna switches (this business was sold to Epcos)
  • MEMS microphone (this business was sold to Knowles)
  • Relative Humidity (RH) and pressure (P) sensors on 140nm platform (this business was sold to ams AG)
Foto
A PICS carrier seen from the top, with an active RF transceiver die flip-chipped on top of it.
Foto
Schematic cross-section of the PICS-active assembly
Back
Home
Contact
Picture
Web design by Margot and Pieter Hooijmans using Weebly.
  • Quick Navigation
  • Home
    • About Pieter Hooijmans
    • About Maximus-R&D
    • Experience >
      • Radar Technology
      • Optical communications
      • Tuners and RF Modules
      • RF IC's
      • Communication Systems
      • Audio and Analogue
      • IC Technology
      • Packaging
    • R&D Processes
    • Services >
      • Client Projects
    • Contact
  • Technology History
    • Piet Hooijmans 1918 - 2006
    • Piet's Home-built Television pt1
    • Piet's Home-built Television pt2
    • EQ40 and EQ80
    • TV Tuner history pt1
    • TV Tuner history pt2
    • TV Tuner history pt3
    • Philips TV remote controls, 1955-1985
    • TV Tuner history pt4
    • TV Tuner history pt5
    • TV Tuner history pt6
    • Digital circuit blocks
    • TRANSDECO
  • Oil Painting
  • Vliegen met Pieter