As described in the EDA360 vision paper, Silicon
Realization represents the creation of
IP blocks, ICs, or systems-on-chip (SoCs) ready for software integration.
But how is it different from EDA as we know it today, what are the challenges,
and what solutions are needed? Sandeep Mehndiratta, group director of solutions
marketing at Cadence, answers these and other questions in the following
Q: The EDA industry
has supported silicon design all along. What's new and different about the
EDA360 Silicon Realization concept?
A: What's different is that the bar has shifted. Today,
engineers have to design a chip with integrated functionality that was formerly
distributed over multiple boards. A single chip may include audio, video, touch
sensors, ambient temperature sensors, and more. Multiple applications are
running at the same time. Then there's a manufacturing aspect - you have to
ensure high yield and make sure the packaging cost is within budget.
If you look at a datasheet for a chip, it talks about
function, power, performance, and unit cost. Our customers have to ensure that
their silicon meets all these goals. But the flows available today are still
legacy design flows based on sequential optimization. Designers first integrate
the functionality, then do mixed-signal integration and verification, then
optimize for power, and then hand off to packaging. There are massive
iterations throughout the design, verification and implementation phases. These
iterations result in cost overruns, schedule overruns, and missed
The promise of Silicon Realization is concurrent, objective
optimization for power, mixed-signal, performance, packaging, and advanced node
enablement. It is based on integrated, interoperable flows.
Q: What are the
requirements for Silicon Realization?
A: Silicon Realization is about design intent, design
abstraction, and design convergence.
Design intent means having an early, unified representation
of the specifications for behavioral,
timing, power, and electrical and physical aspects of the design. By capturing
intent early, you address issues up front and reduce iterations. For example, common
constraints between analog and digital flows cut implementation time and reduce
verification cycles. Compliance and metric driven verification deliver
predictable schedules and improve quality, while ensuring that unified power
intent is enforced from system analysis to physical implementation.
Design abstraction is about using higher level views of the
design's function, timing and even physical behavior to do early trade-off
analyses, accelerate validation, and reduce iterations. It is about using wreal [wired real] models for analog blocks
to run mixed-signal verification at digital speeds. It is about using rapid
prototyping for digital architecture exploration and floorplanning.
The EDA360 vision talks about starting with the applications
and operating system and moving down. Design convergence is about combining
top-down design and bottom-up methodologies to enable early analysis and
optimization, and avoid issues later in the cycle. Convergence is enabled
through early power estimation and analysis, in-design electrical and physical sign-off,
and early planning. IC/package co-design is a convergence story because you're
abstracting and you're making a decision earlier in the design cycle.
Q: How is Cadence
delivering, or promising to deliver, on these approaches?
A: There are many proof points in the Cadence flow that
satisfy requirements for design intent, abstraction, and convergence. These
include multi-objective, physically-aware optimization; power-aware IP reuse
with macromodeling; metric-driven verification; accelerated mixed-signal
verification; ECO automation; mixed-signal interoperability; IC/package
co-design; and design exploration and planning.
To take a specific example, Virtuoso 6.1.4 and Encounter
Digital Implementation System 9.1 have common rules and models based on the
OpenAccess infrastructure, enabling digital designers to get a gray box
representation of an analog model and vice versa.
Realization has a strong mixed-signal message. Mixed-signal design has been
around for years. What's new?
A: Customer challenges around mixed-signal design,
verification, and implementation are accelerating and expanding to virtually
every chip they are designing. Any time a chip does real-world data handling,
designers need to do mixed-signal integration and verification. Many IP
building block components today are mixed-signal in nature.
Our customers not only have to worry about the functionality
and verification of devices -- they have to worry about integrating the analog
and digital interfaces. And there are mixed-signal challenges around advanced
process nodes. Driven by integration cost factors, RF blocks will be put on
smaller nodes, where the chances of failure will be higher. Mixed-signal
challenges are a common cause of silicon failures for IPs, ICs and SoCs.
Q: The EDA360 vision paper talks a great deal about integrators. It's
clear that Silicon Realization is aimed at IP creators, but does it also apply
A: Absolutely. There is an industry shift from pure creation
to creation and integration. The goal is to differentiate in core technologies and
to reuse or source the non-differentiating parts. For integrators to do that
within the right budgets, and be able to integrate and deliver solutions on
time, they need new methodologies and IP that is ready for integration.
Look at power, for example. Lots of advanced techniques are
being used in power optimization for IP. So you cannot black-box the IP just
for function and timing - you need power-aware optimization. Similarly, if you
wait for the entire SPICE representation of an analog component to be ready for
mixed-signal integration, you may have issues with bugs that are found too
What integrators need is a representation that lets them
integrate the design in multiple contexts. They need the ability to recognize
multiple facets of the design intent including function, power, and timing.
They need abstractions of blocks at a level where they can do concurrent design
and handle ECO changes, and they need methodologies that allow them to do accelerated,
parallel design verification and implementation. Silicon Realization addresses
Q: You talked about
moving to a higher level of design abstraction. The move to TLM-based IP
creation and integration is an obvious example for digital, but what about
analog and mixed-signal design and verification?
A: Abstraction means the ability to represent the attributes
of the design in a consistent form while it is still tied to the lowest level
of granularity at which the design is defined. In the analog world,
methodologies are evolving for abstraction. One example is wreal models. These models let you model voltages and currents in a
digital context. You're abstracting the design representation and still doing
accurate representations of the functional and electrical behavior of the
circuit, without waiting for a full SPICE representation.
Another example of abstraction is capturing power intent of
IP blocks using macromodels, enabling power-aware integration and verification.
You can represent the embedded power behavior of the circuit using CPF [Common
Power Format], which lets users model power supplies, multiple domains, and
circuit inputs and outputs. Another example of abstraction is the ability to
extract a full timing model from Virtuoso for an analog/mixed-signal block, so
it can be used in full-chip static timing analysis.
Q: Verification has
become a huge challenge for IC design teams. From a Silicon Realization
perspective, what capabilities are needed?
A: There are two aspects to this. One is that functionality
is expanding, and mixed-signal is a manifestation of that. We need to scale
functional verification to apply techniques around metric-driven verification
to the mixed-signal world, and do that without giving up performance. The other
aspect is power. An industry study showed that 87 percent of design respins at
40 nm involved leakage problems. Power is becoming a central aspect of
verification, and what's needed is a unified intent representation that starts
at the IP level and flows into integration and verification.
Q: How is Cadence
bringing the Silicon Realization message into the Design Automation Conference?
We will be demonstrating Silicon Realization solutions at
the Cadence booth. We are also planning a Silicon
Realization lunch panel Tuesday, June 15, from 11:30 am to 1:00 pm. A panel
moderated by Cadence CMO John Bruggeman will include presenters from Cadence,
Bridge, and Qualcomm.
They will discuss how to go from concept to silicon as quickly and cheaply as
possible, while retaining the high quality that end customers demand.