IEEE Solid-State Circuits Magazine - Winter 2016 - 40

VCC
LC

OUT+

RB1

Q1
vAMPLIT

OUT-
LB
Q2

RB2

D2
RV

ICS

Vfreq

ICS

Figure 1: The differential Colpitts VCO topology [3].

Leonard Dauphinee at the IEEE International Solid-State Circuits Conference (ISSCC) in 1997 [1]. Copeland
and Colpitts are a perfect fit. Few
people are aware, and I am sure that
not even Miles knew at the time,
that although universally known as
an American scientist and engineer,
Edwin Colpitts was born, raised,
educated, and buried in New Brunswick, Canada.

Nortel University
I first met Miles in the fall of 1994
when, after my Ph.D. studies, I joined
Peter Schvan's R&D group at Northern
Telecom (Nortel) in Ottawa, Canada.
The group consisted of several fulltime Nortel researchers, consulting
professors from Carleton University,
and their graduate students. It was a
hotbed of RF and fiber-optic silicon
(Si) IC research and development,
covering RF Si-CMOS, Si and silicongermanium (SiGe) bipolar and Si-BiCMOS process development, modeling
of active and passive devices, RF,
10-Gb/s and 40-Gb/s fiber-optic
IC demonstrators. No wonder that
one of the former members of the

W I N T E R 2 0 16

group called it, many years later,
"Nortel University."
Miles and his graduate students
were focusing on developing the Siintegrated inductor and transformer
modeling program GEMCAP2 and
on demonstrating the first Si and
SiGe bipolar RF benchmark circuits
in the 1-5 GHz range. Since most RF
and microwave circuits were single
ended in those days, Miles and Ph.D.
student Leonard Dauphinee were
searching for differential topologies that could be used at RF. It was
in 1995-1996 that they proposed
the first differential LC varactortuned version of the 80-year-old
classic Colpitts oscillator topology.
A 1.5-GHz voltage-controlled oscillator (VCO) was integrated in Nortel's 11-GHz Si-BiCMOS process and
presented by Leonard at the ISSCC
in February, 1997.
As illustrated in Figure 1, this
circuit features built-in isolation
between the tank resonator, formed
by L B, C E and the varactor diode
D 1, and the load impedance consisting of R C , L C . Compared to
the alternative cross-coupled VCO

IEEE SOLID-STATE CIRCUITS MAGAZINE

topology that had been revived two
years earlier from its first pre-World
War II vacuum-tube incarnation and
adapted for Si integration, it has
an extra degree of design freedom
through C E that allows for phase
noise minimization.
Although the design equations
for the oscillation frequency and
the onset of oscillation were well
understood, modeling the parasitic
capacitance and Si substrate losses
of inductors was rather rudimentary
and continues to be a problem even
today. Leonard had to run several
test chips through the fab, intentionally setting the VCO frequency
about 10% higher than the desired
value. This prompted him to glumly
and sheepishly coin the phrase: "the
fab is my simulator."
Second, and this was a provocation for Miles, unlike low-noise amplifiers and mixers, it was surprisingly
challenging to precisely predict its
output swing and phase noise analytically and thus to develop an algorithmic design methodology. Many
design iterations using large-signal
and phase noise periodic steadystate simulations were needed to
optimize an oscillator.
Nevertheless, as a testament to
its robustness and outstanding performance, the differential Colpitts
VCO topology was shortly scaled
thereafter to mm-wave frequencies
[2] and adopted in high-end, low-jitter, fiber-optic transceiver products
by Nortel and Ciena, 10-Gb/s SONET
and 10-Gb Ethernet transceivers
by Quake Technologies [3], in Infineon's 77-GHz automotive radar
transceiver, and Peraso's 60-GHz and
radio transceiver products.
By 1997, now retired from Carleton, Miles began to consult three

Table of Contents for the Digital Edition of IEEE Solid-State Circuits Magazine - Winter 2016