What’s Next After Moore’s Law?
The Semi Startup Renaissance Part 1
The Semiconductor Startup Renaissance: Excitement and investment in the semiconductor space has dramatically increased in the past few years. Drawing on our deep knowledge of the sector and extensive portfolio of successful semiconductor investments at Celesta Capital, this article series explores the major themes driving renewed interest in chip innovation.
Semiconductor startups and venture capital have been intertwined from their inception. When Art Rock arranged the financing necessary for Gordon Moore, Robert Noyce, and six of their Shockley Semiconductor colleagues to quit and create Fairchild Semiconductor, it ushered in an explosion of new semiconductor companies so impactful it forever earned the Bay Area the nickname Silicon Valley.
Yet, since the turn of the millennium, VC attention has been dominated by software. Much of this was driven by the tremendous success of many early software startups, as well as their perceived low overhead-high margin business models. But semiconductor startups have also been victims of their own success – specifically, the success of Moore’s law.
Moore’s law predicts that the number of transistors on a chip will double every two years. So long as Moore’s law held true, users could count on shrinking transistors to deliver decades of exponential growth in compute power and efficiency. And in fact Moore’s law did continue to hold true for decades, which has had a chilling effect on semiconductor startups in three ways:
Architecture: New hardware architectures require new software. Even if a new architecture promises performance gains when it comes out in the future, it is difficult to justify rewriting software when you know CPUs will continue to improve exponentially.
Specialization: Specialized chips for an application or market can deliver performance gains. However, those gains are quickly eroded when the entire industry is pushing hard to constantly improve general purpose processors every two years.
Systems: Engineers can eke out performance gains through system-level optimization and architecture choices. The effort to do so must always be weighed against the knowledge that Moore’s law will give performance gains without effort to those who wait.
For a long time, designing a product to take advantage of the next generation of Moore’s law was enough to delight users.
Today, however, this dynamic has begun to shift as the pace of Moore’s law has slowed considerably. You can only shrink a transistor so much until you must pursue other pathways to improve performance. There is little consensus on the exact timing of the demise of Moore’s law (or even whether it has truly ended), but most will agree it began slowing in the 2010s. Nvidia CEO Jensen Huang famously declared onstage at CES in 2019 that Moore’s law was “dead”.
Regardless, consumers are accustomed to the drumbeat of regular performance improvements. If we can’t achieve this from transistor shrink, then, from where? Where Moore’s law has stumbled, chip startups are off to the races.
The undercurrent of major trends driving semiconductor startups is the need to improve performance through alternative means and venture investors have started paying attention and writing checks.
Suddenly, what was old is new again and innovations in architecture, specialized chips, and system-level optimization are at the forefront. Startups are embracing unorthodox architectures that, among other things, try to reduce the bottleneck accessing memory. Companies are being built around application-specific processors in large markets. Two separate networking startups, Astera Labs and Celesta portfolio company Credo, have had successful public debuts in the last few years.
There is no time like the present for semiconductor startups.