History

Work

  • 2009-2019: IBM's T.J. Watson Research Lab — Research Staff Member
  • 2008-2009: University of Michigan — Postdoctoral Researcher
  • 1998-2002: Analog Devices Inc., Ireland — Design Engineer

Education

  • 2003-2008: Ph.D., University of Michigan
  • 1994-1998: B.E., National University of Ireland, Cork (UCC)

Topics

  • Self-healing PLLs
  • Fractional-N noise cancellation
  • Varactor folding
  • Flicker noise suppression in VCOs
  • Software for phase arrays

IBM's mm-Wave group's big ICs


60GHz Radio

28GHz 32 element phased array

Self-healing PLLs




Ferriss, M.; Plouchart, J.-O.; Natarajan, A.; Rylyakov, A.; Parker, B.;
Babakhani, A.; Yaldiz, S.; Sadhu, B.; Valdes-Garcia, Alberto; Tierno, J.; Friedman, D.,
"An integral path self-calibration scheme for a 20.1–26.7GHz dual-loop PLL in 32nm SOI CMOS,"
VLSI Circuits (VLSIC), 2012.

Single and dual path PLLs


  • Split control path into proportional and integral path e.g: [Craninckx, JSSC 1998]
  • Requires two charge-pumps and two VCO varactors

    Why use this architecture?

VCO small signal gain variation


Gain variation affects on PLL phase noise


Prior-art bandwidth correction


Effects of non-linearity and offsets


  • Saturation corrupts measurement when step is large
  • Offset corrupts measurement when step is small

Time-to-crossover in presence of offsets


Real measurement has offsets caused by: Charge pump current mismatch, capacitor leakage, BB-PFD offset, etc.

PLL response to a phase step, with no proportional path


  • Saturation corrupts measurement when step is large
  • Offset corrupts measurement when step is small

Transfer function before/after calibration


Calibration system stabilizes transfer function

Phase noise before/after calibration


Stabilized transfer function leads to phase noise uniformity

Wafer level test


Measured jitter peaking (in dB) before/after calibration. Tested every die on a 300mm wafer at 25GHz

Improved uniformity shown across wafer

Parametric phase noise healing


Noise @10MHz = f( VCO fnom, Amplitude, Bias Voltage and Current, VCC, Band )

Methodology described in: S. Yaldiz, V. Calayir, X. Li, L. Pileggi, A. Natarajan, M. Ferriss and J. Tierno, “Indirect Phase Noise Sensing for Self-Healing Voltage Controlled Oscillators,” IEEE CICC, Mar. 2011.

Fractional-N noise cancellation




Ferriss, M.; Sadhu, B.; Rylyakov, A.; Ainspan, H.; Friedman, D.,
"A 13.1-to-28GHz fractional-N PLL in 32nm SOI CMOS with a ΔΣ noise-cancellation scheme,"
IEEE International Solid- State Circuits Conference (ISSCC), 2015.

Objective


Hybrid PLL Architecture


Baseline integer-N Hybrid PLL from [Ferriss, et al., JSSC 2014]

Classic Fractional-N PLL


Classic $\Delta\Sigma$ Problem


  • Divider’s $\Delta\Sigma$ dithers PLL’s feedback (FB) clock
  • $\Delta\Sigma$ noise contributes to PLL’s phase noise

Time-based Cancellation Concept



  • Digital-to-time based delay scheme used on analog path
    • Utilizes loop filter’s low pass response to remove delay path quantization noise -> Don’t need fine quantization step.

Componets of delay generation


Delay Controller Architecture


Controller’s function is to:
(1) Background calibrate DTC gain, (2) Shape $\Delta$T quantization error, and (3) Shape the mismatch error

Delay Controller Operation Simulation


Integral Path Introduction


How do we interpret early/late information?

Integral Path Prediction Concept


Integral Path Prediction Circuit


  • Integral path operates from single BB phase detector
    • Works in parallel with the analog proportional path
  • Ignores BB results that match prediction

Dual D/VCO Scheme


Die Photo with Layout


Measured Phase Noise and Output Spectrum


Demonstration of DTC Background Cal


Background calibration engine finds the optimum setting

Dual D/VCO Tuning Characteristic


Transient Locking


Response to step in division ratio measured with sampling scope.
Measured with different integral path gain settings (green, red), and with direct loading of digital control word (blue).

Varactor Folding




Ferriss, M.; Sadhu, B.; Rylyakov, A.; Ainspan, H.; Friedman, D.,
“A 12-to-26GHz fractional-N PLL with dual continuous tuning LC-D/VCOs,"
IEEE International Solid-State Circuits Conference (ISSCC), 2016.

Oscillator Frequency Control


Motivation


  • Objective: Remove coarse bands of CMOS oscillators
  • Utilize entire tuning range to generate frequency chirps

Digital control of a varactor


Trade off between SD noise and Gain/Tuning range

Extending the range: varactor folding



Similar concept used in [P. Wang, et al., JSSC 2009]

Comparing typical tuning mechanisms



Can we use MOM+switch for continuous tuning?

Varactors vs MOM+switch (Cap)


Varactors vs MOM+switch (Gain)


Varactors vs MOM+switch (Q)


Q degradation versus # of elements


Single band


*Connections for va0 bias omitted for simplicity

Two bands with offset


*Connections for va0 bias, vb0 bias omitted for simplicity

Two bands and two thresholds


*Connections for va0 bias, vb0 bias omitted for simplicity

Varactor gain versus frequency


In [1]:
%%HTML
<div align="middle"><video width="100%" controls><source src="Inkscape/Folding/media1b.m4v" type="video/mp4"></video>

Patent contains a few alternative configurations


The VCO


PLL architecture


32nm CMOS prototype


Measured tuning range


Measured phase noise


VCO’s FOM = 181dBc/Hz, FOMT = 188dBc/Hz @10MHz offset

Transient frequency measured with 80GHz oscilloscope


In [3]:
%%HTML
<div align="middle"><video width="100%" controls> <source src="Inkscape/Folding/media2b.m4v" type="video/mp4"></video>

Chirp generation comparison


Spectrum when generating chirps


Flicker noise suppression in VCOs




Ferriss, M.; Sadhu, B.; Friedman, D.;
"A Gradient Descent Bias Optimizer for Oscillator Phase Noise Reduction Demonstrated in 45nm and 32nm SOI CMOS,"
IEEE Radio Frequency Integrated Circuits Symposium (RFIC), 2018.

Overview of architecture


Oscillator sensitivity to noise


Transconductance


Transconductor with odd I-V characteristic => no up-conversion from DC to phase noise. [Pepe, Andreani TCAS I, 2017]

Transient behavioral simulation


Die photos: (a) 45nm, (b) 32nm.


Measured frequency sensitivity to bias voltage


45nm phase noise (top) and current (bottom). Squares are with biasing scheme enabled.


45nm, bias scheme enabled, disabled, and swept


32nm with tail inductor, bias scheme enabled, disabled


32nm output spectrum


Software for phased arrays




Sadhu, B.; Paidimarri, A.; Ferriss, M.; Yeck, M.; Gu, X.; Valdes-Garcia, A.,
"A Software-Defined Phased Array Radio with mmWave to Software Vertical Stack Integration for 5G Experimentation,"
2018 IEEE/MTT-S International Microwave Symposium - (IMS), 2018.

Sadhu, B.; Tousi, Y.; Hallin, J.; Sahl, S.; Reynolds, S.; Renström, Ö.; Sjögren, K.; Haapalahti, O.; Mazor, N.; Bokinge, B.; Weibull, G.; Bengtsson H.; Carlinger, A.; Westesson E., Thillberg, J.; Rexberg, L.; Yeck M. , Gu X.; Ferriss M.; Liu, D.; Friedman, D.; Valdes-Garcia, A.,
"A 28-GHz 32-Element TRX Phased-Array IC With Concurrent Dual-Polarized Operation and Orthogonal Phase and Gain Control for 5G Communications”
IEEE Journal of Solid-State Circuits, 2017. Awarded JSSC 2017 Best Paper.

The 5G module


Photograph of the SDPAR system.

Simulating non-idealities


Beam optimization (Software)



An optimization experiment using the SDPAR software emulator where the optimizer creates a beam pattern to improve signal to interference ratio starting from an arbitrary beam.

Beam optimization (Hardware)


The End