Key Moments
Is Moore's Law Finally Dead?
Sabine HossenfelderKey Moments
Moore's Law isn't dead, but its original form faces limits. New tech like specialization and 3D stacking are pushing progress.
Key Insights
Moore's Law, stating transistor count doubles roughly every two years, has driven computing power but faces physical and economic limits.
Miniaturization challenges include reaching atomic scales, managing heat, and quantum effects like electron tunneling.
The cost and complexity of chip manufacturing have led to industry consolidation, with fewer companies focusing on cutting-edge transistors.
Specialization (heterogeneous computing) and advanced fabrication methods like Extreme Ultraviolet Lithography are current strategies to enhance performance.
3D stacking of transistors (chiplets) offers increased density but faces significant thermal management challenges.
Emerging technologies like graphene, carbon nanotubes, quantum computing, photonics, and spintronics are being explored in labs but are not yet commercially viable.
THE ORIGINS AND LIMITS OF MOORE'S LAW
Moore's Law, an observation by Gordon Moore in 1965, predicted that the number of transistors on a microchip would double approximately every two years. This exponential growth, driven by transistor miniaturization, has led to remarkable increases in computing power since the first microchips. However, this progress is hitting physical and economic walls. Transistors are approaching atomic size, making further miniaturization impossible as an atom is not a transistor. Additionally, increased transistor density generates significant heat and introduces quantum effects like electron tunneling, which can disrupt performance.
MANUFACTURING CHALLENGES AND INDUSTRY SHIFTS
Producing modern microchips has become incredibly complex, involving hundreds of precise steps. As transistors shrink to the nanometer scale, they become more susceptible to defects, necessitating advanced fabrication and monitoring techniques like e-beam screening and optical inspection. These additional steps significantly increase production costs. The enormous investment required for cutting-edge chip factories (billions of dollars) has led to industry consolidation. Many companies have shifted to producing older, cost-effective 'legacy' chips, with only a few major players like TSMC, Samsung, and Intel pushing the boundaries of miniaturization.
HETEROGENEOUS COMPUTING AND ADVANCED FABRICATION
To circumvent traditional miniaturization limits, the industry is embracing specialization through heterogeneous computing. This approach involves using multiple types of processors, each optimized for specific tasks, rather than relying on a single, general-purpose CPU. Examples include GPUs for parallel processing and NPUs for machine learning. This strategy enhances efficiency and speed without requiring a radical departure from existing production technology, making it a cost-effective solution. Concurrently, advanced fabrication techniques like Extreme Ultraviolet (EUV) Lithography are being implemented. EUV uses shorter wavelengths of light to etch smaller features onto silicon wafers, enabling more intricate chip designs.
THE PROMISE OF 3D TRANSISTOR STACKING
A significant frontier in semiconductor advancement is moving into the third dimension. While transistors have evolved from flat designs to 3D structures like FinFETs for improved performance, the next major step is stacking these transistors on top of each other. Technologies like Intel's Foveros Direct and TSMC's Chip-on-Wafer aim to pack transistors volumetrically rather than just across an area. This 3D stacking can dramatically increase transistor density. However, a major hurdle is heat dissipation; packing transistors tightly makes it difficult to remove the generated heat, requiring innovative cooling solutions like channels within the chip stack itself.
EXPLORING NEW MATERIALS FOR MINIATURIZATION
Beyond silicon, researchers are investigating alternative new materials that could facilitate smaller and more efficient transistors. Graphene, a single layer of carbon atoms, offers excellent heat conductivity but struggles with its 'band gap,' hindering its ability to act as a switch. Carbon nanotubes, rolled-up graphene structures, show promise as they can be made conductive or semiconducting, but commercial-scale production remains a challenge. Other materials like gallium nitride, germanium-tin alloys, and molybdenum disulfide are also being explored in laboratories, each with potential advantages, though none are close to market readiness.
NOVEL COMPUTING PARADIGMS AND FUTURE OUTLOOK
The pursuit of more advanced computing also involves exploring entirely new paradigms beyond traditional transistor-based operations. Quantum computing, utilizing quantum bits (qubits), and photonic computing, using light, are active research areas. Quantum dots are being investigated as a way to leverage quantum effects for computation. Photonic computing benefits from photons being small and fast, but their lack of self-interaction poses challenges. Spintronics, which uses the spin of electrons, requires significantly less power but often necessitates extremely low temperatures, although some promising room-temperature solutions are emerging. While many of these laboratory-stage technologies face scaling and cost barriers, they represent the ongoing spirit of innovation pushing the boundaries of computing power.
Mentioned in This Episode
●Supplements
●Products
●Software & Apps
●Tools
●Companies
●Organizations
●Concepts
●People Referenced
Navigating the Future of Computing Power
Practical takeaways from this episode
Do This
Avoid This
Common Questions
Moore's Law is an observation made by Gordon Moore in 1965 stating that the number of transistors on a microchip roughly doubles every two years. This trend has historically driven the increase in computing power.
Topics
Mentioned in this video
One of the remaining four major chip manufacturers pushing miniaturization and integrating EUV lithography.
Has taken an interest in carbon nanotube transistors.
An early personal computer with 128 kilobytes of RAM and a resolution of 320x200 pixels, mentioned as a point of comparison to modern devices.
The pioneer of Graphics Processing Units (GPUs), which are an example of specialized processors used in heterogeneous computing.
An industry group that predicted in 2015 that shrinking transistors in microprocessors might not make financial sense after 2021.
Graphics Processing Unit, pioneered by NVIDIA, known for parallel processing and now used for various tasks beyond graphics.
Tiny tubes rolled from graphene, which can be conducting or semiconducting and used to build tiny transistors. Production is difficult and expensive.
A material combined with graphene by researchers to achieve switching of electron spins at room temperature for spintronics.
A password keeper offered by the same company as NordVPN.
A single layer of carbon atoms with potential applications in transistors due to its efficient heat dissipation, though it has issues with band gap and commercial scalability.
A chip manufacturer involved in the debate about the current state of Moore's Law and developing new technologies like Foveros Direct.
Co-founder of Intel who observed the doubling of transistors on a microchip roughly every two years, leading to Moore's Law.
A strategy that uses multiple types of processors, each for specific tasks, to increase computer speed and efficiency.
Intel's fourth iteration of 3D stacking technology, significantly increasing transistor density.
A material being researched as a potential alternative to silicon for transistors, showing promise in laboratories.
A material being researched as a potential alternative to silicon for transistors.
A computing paradigm that uses light (photons) to perform calculations, offering speed advantages but facing challenges with photon interaction.
A type of 3D transistor design where the conducting channel stands out, improving cooling, reducing power consumption, and speeding up operations.
A grant program created by DARPA to support research on thermal management of 3D transistor stacks.
A type of crystal studied for its potential as a spintronics transistor that could operate at room temperature.
A secure file storage and sharing platform offered by the same company as NordVPN.
The observation that the number of transistors on a microchip doubles approximately every two years, driving computing power increases.
A new transistor technology that turns the channels by 90 degrees so the gate wraps around them.
A material being researched as a potential alternative to silicon for transistors.
Nanoscale sized structures in semiconductors that embrace quantum properties and are being explored for use in quantum computing transistors.
Taiwan Semiconductor Manufacturing Company, a major chip manufacturer that is one of the few left pushing miniaturization and is developing 3D stacking technology.
An example of a chip that uses heterogeneous computing, integrating CPUs, GPUs, and other processing units.
Researchers at this institution repurposed a commercial silicon manufacturing plant to create carbon nanotube transistors.
A material being researched as a potential alternative to silicon for transistors.
A hardware option that uses the spin of electrons as the basic unit of calculation, offering lower power consumption and higher speed, but often requiring extreme cooling.
One of the remaining four major chip manufacturers pushing miniaturization.
Central Processing Unit, one of the common types of processing units found in computers, often paired with GPUs in heterogeneous computing.
Neural Processing Unit, a type of processor specialized for machine learning tasks, used in heterogeneous computing.
An advanced lithography technique using ultraviolet light with a short wavelength (13.5 nm) to etch smaller features onto silicon wafers, crucial for modern chip production.
The US Defense Advanced Research Projects Agency created the Minitherms3D grant program to support research on thermal management of 3D stacks.
A type of computer that uses quantum bits (qubits) which can act as bits, offering a different approach to computation.
A VPN service that sponsors the video, offering secure internet connections, privacy, and threat protection.
More from Sabine Hossenfelder
View all 27 summaries
7 minBreakthrough In Data Storage Could Store Your Photos for 10000 Years
7 minThe Simulation Hypothesis Gets Scientific Backing
7 minSurprise! Milky Way Might Not Have a Black Hole After All
7 minThe First Moon Landing Wasn’t Apollo — And We Just Found It
Found this useful? Build your knowledge library
Get AI-powered summaries of any YouTube video, podcast, or article in seconds. Save them to your personal pods and access them anytime.
Try Summify free