Anda di halaman 1dari 2

HARDWARE EMULATION ANSWERS

AI/ML VERIFICATION NEEDS


With AI/ML chip designs containing between 5 billion and 10 billion gates, design verification using
hardware emulation is the answer, although not all hardware emulators are the same

By Lauro Rizzatti Artificial intelligence/machine learning (AI/ back to what Moore’s Law is supposed to do.
ML) is the ultimate hot topic of 2019, taking
AI/ML is imposing tremendous capacity
hold of chip design and the semiconductor
needs on the market, and these designs also
industry’s imagination and not letting go
present a challenge to the chip verification
— for good reason. Close to 1,000 startups
market. While the processor may be a rela-
in China are in the AI/ML space, along with
tively simple design, the AI/ML space needs
quite a few in the U.S.
to deploy many of them and scale quickly.
Moore’s Law is being rejuvenated by AI/ML Another consideration is software that must
after capacity fears slowed it down. Later this be verified along with hardware.
year, estimates put AI/ML design capacity
between 5 billion and 10 billion gates — huge Hardware emulation is the answer,
capacity for a single design. Designs will go although not all hardware emulators
are the same.

One high-profile startup in this market


recently announced adoption of hardware
emulation for its large AI/ML chip design
verification environment. It selected a hard-
ware emulation platform with the largest
design capacity commercially available and
with a rigorous roadmap for the future. The
vendor’s sophisticated and experienced
engineers designed the architecture and the
chip in-house using their progressive knowl-
edge. The environment has its own operating
system and a complete software solution. It
Artificial intelligence (AI) and machine learning (ML) offers scalability, capacity, throughput, and
(SOURCE: PIXABAY.COM)
emulation platform itself, which, in this case,
is scalable and supports a virtual system
environment because there is no legacy ICE
environment that needs to be supported.

Commercially available and competitive


hardware emulation platform offerings
are different depending on the vendor, so
designers need to perform thorough eval-
Artificial intelligence (AI) and machine learning (ML) uations when it comes to selecting a hard-
(SOURCE: PIXABAY.COM) ware emulation platform for use in an AI/
a deterministic verification environment. ML design verification environment. Some
For example, the one-box emulator provid- approaches may appear to be similar, but
ed about 2.5 billion gate capacity when it they may have fundamental differences
was introduced two years ago. Users today “under the hood.” For example, one hard-
achieve a capacity of 5 billion gates. As ware emulation vendor designs its own chip,
capacity needs continue to soar for AI/ML but power is a problem as the architecture
designs and other massively complex chips, doesn’t scale well with power as a specifi-
the hardware emulation vendor pledges that cation and its roadmap is flawed. Another
it will meet the demand with up to 10 billion vendor uses FPGA chips to map the design
gate design capacity. In fact, its hardware onto the emulator. This approach has good
emulator is deployed now at user sites, performance, but design compilation is diffi-
enabling determinism to verify large systems. cult, as is determinism, and this solution also
provides poor visibility into the design.
Of course, designers at this startup could
have used another emulator if their chip AI/ML is a vibrant topic in the semiconductor
had been smaller and didn’t need to scale industry, spawning great new product ideas
or if they didn’t require visibility into the and a great deal of investment. Some hard-
design. As designers start to scale to larger ware emulation vendors are well-positioned
chips, however, they need to run hardware to tackle the verification challenge that these
and software and probe with a high amount new chip designs present, others not so much,
of visibility. They also need a deterministic so it behooves AI/ML design and verification
environment because an AI/ML design is teams to do their homework before spending
deterministic, but this is not an option avail- a lot of money on a system that doesn’t sup-
able with all hardware emulation platforms. port the capacities and provide the capabili-
Another deciding factor is the hardware ties demanded by their AI/ML devices.
2

All contents are Copyright © 2019 by AspenCore, Inc. All Rights Reserved.

Anda mungkin juga menyukai