No one would doubt that Intel's Core
2 Duo is perhaps one of the most hyped and anticipated processor launches since
the K8. The processor, expected to start shipping on
July 23, 2006, is Intel's first major transition in desktop
microarchitecture since the Willamette core was introduced in
Conroe and its derivatives are a step away from Intel's former
flagship NetBurst, but even these processors are a bit of a dying breed: during
Intel's shift to 45nm, the company will no longer focus on derived
microprocessor cores in favor of refined unified core architectures. For
example, later this year Intel is expected to release Allendale, a dual core Conroe-based
processor with only 2MB of L2 cache. From what Intel has stated
from forward looking statements and roadmaps, this sort of practice will
One possibility as to how this will work will be in the way Intel packages its
processors. For example Intel's Presler is essentially a Cedar
Mill processor with two dice coupled on the same package. Likewise,
Intel's Kentsfield processor is two Conroe dice coupled on a
single package to make a quad-core processor. Even though your server and
desktop processor will have the same processor core on Intel's future
architectures, the logical way to differentiate the role of these CPUs will be
the number of dice per package.
Intel announced the majority of its long-term architecture plans in April of
this year. During this same announcement, Intel detailed that the Core
microarchitecture would survive into 2008 with the 45nm Penryn core.
Penryn, which will be based on Intel's lithography process known as P1266, is a 45nm unified
core set for launch in 2007 that is also expected to stay into production into
2008. Intel has already produced SRAM samples for 45nm CPUs, as
demonstrated in the lithography shot on the right. Aside from the process
shrink on Penryn, the major divergence in design from Conroe is
the new material design. With P1266, Intel shifts away from Silicon
Dioxide gate dielectrics -- a process the company has used since the mid-90s --
to High-k dielectrics. With new dielectric techniques, the company will
also revamp its gate electrodes to metal instead of Polysilicon
derivatives. The last major materials change of such magnitude occurred
when Intel moved from Silicon to Strained Silicon in 2002, which is still
slated for use in P1266 and beyond.
According to Intel's long term roadmap, Penryn will become the last
"Core" (NGMA) microarchitecture, but it will not become the last 45nm
generation. A new architecture, based on the Nehalem core, will
make an appearance in 2008. This Core successor, dubbed by many as NGMA2,
will essentially take all of the 45nm manufacturing principles of Penryn and
apply them to the totally new Nehalem architecture. This transition will
be very similar to the P6+ transition with Yonah from Dothan to Conroe.
If P1266 was radically different from the 65nm Core CPU process, P1268 takes what is
normal on 45nm and throws it out the window. The first P1268
processor will use a CPU codename that Intel has publically announced as Nehalem-C. Nehalem-C
is still based on the Nehalem (NGMA2) microarchitecture, but is a die
shrink from 45nm to 32nm. P1268 employs all of the neat manufacturing
techniques found on P1266, but may use a totally different lithography process
called EUV, or Extreme Ultraviolet. Although it sounds trivial to swap a
traditional DUV lithography setup with an EUV one, there is a fundamental
problem with EUV that prevents foundries from doing so; the EUV wavelength is
so small at 13.4nm that virtually every material absorbs the wave including the
lenses and atoms in the air. EUV must be done in a vacuum with reflective
surfaces instead of lenses to focus and redirect the lithography.
However, EUV isn't the only radically new tool in Intel's arsenal for
2009. According to Intel's forward statements, Nehalem-C is
succeeded by Gesher, the third in Intel's next generation
microarchitecture designs uniquely dubbed NGMA3. Intel hinted last week
during the 2006 VLSI symposium that it will start using tri-gate transistors on
its 32 or 22nm production processors. Since the 1950s, transistors have
been strictly planar designs, with gates that lay flat across the
substrate. A tri-gate design is unique in the fact that a single gate is
stacked on top of two vertical gates allowing for essentially three times the
surface area for electrons to travel. Whether or not these tri-gates will
appear on Gesher or its 22nm derivative have not been announced.
Even beyond Gesher and its 22nm derivative, Intel has a lot of
technologies in its roadmap. One technology proposed to replace tri-gates
is by using silicon nanowires surrounded by a metal gate to increase surface
area even more. Intel laboratories have also experimented with carbon nanotubes
a mere 1.4nm in diameter as a future transistor material, yet these materials
won't show up until 2013 at the earliest.
If anything, here are the big points to keep in mind for Life After Conroe:
quote: Another point is that Intel already has its unified architecture.
quote: Ah yes, Dempsey and Woodcrest are totally based off the same core. Not to mention Yonah and Presler. How about Conroe and Montecito?
quote: Intel will unify its processor architectures -- no more Merom/Conroe/Woodcrest derivatives. The processor core that you use in your server will essentially be the same chip you use in your notebook.
quote: No you moron,