The Rise of GaN

Have you ever looked for a power brick on Amazon and seen one of those tiny Anker branded power bricks that can charge at 80w and it is smaller than your Macbook charger? If you are anything like me, you are probably amazed at the fact that this tiny thing is able to withstand all that power. Just around a decade ago, a brick that size was a iPhone 5w USB-A plug — far from being able to charge your laptop.

The reason for these bricks getting smaller is because of GaN, or Gallium Nitride. GaN is a power semiconductor with advantages over the traditional silicon-based chips. Advantages such as faster switching, higher efficiency, less heat, and higher power density. A lot of mumbo jumbo, but we'll explain these in a bit.

For now just know that this is super exciting technology here.

But ☝️🤓, before we begin, I think we should explain some terms technical terms here are: like semiconductor and more specifically a power semiconductor is since these terms will be used often.

A semiconductor is a material (silicon, GaN) used to make electronic components. More specifically, it is a type of material that can conduct electricity under some conditions but not others making it ideal for controlling electrical signals.

A power semiconductor specifically handles high voltage and high current, and is used in power conversion, switching, and amplification. These types of semiconductors are found in chargers, power supplies, data centers, electric cars, etc.

An integrated circuit (IC) is a small chip that can perform functions depending on the type of IC like data storage, etc. These are made from a semiconductor material.

For direct current (DC), energy flows in one straight direction, kind of like water flowing through a hose. Batteries give DC power. An alternating current (AC) is where electricity changes direction rapidly, kind of like water sloshing in and out of a bathub. AC power comes wall outlets, or Spider-Man's Electro (or is he DC? 🥁 ).

Silicon's Reign

For a very long time, silicon has been at the top of semiconductors. All of those smart computers we carry around contain this material — phones, tablets, computers, smart watches, electric cars, etc. It is the foundation of modern electronics.

Silicon is abundant, inexpensive, and has excellent semiconductor properties. The entire semiconductor industry — from logic chips (CPUs) to power transistors — was built around silicon. In power electronics, silicon-based devices have powered everything from laptop chargers to industrial motors.

Silicon is still used in many applications where cost is sensitive, and performance demand is moderate. Add this to mature manufacturing to keep prices ultra-low, ultra-high volume markets like consumer tech benefit from its massive supply chain.

So Why Are We Talking About GaN?

We are talking about GaN (Gallium Nitride) because the world is changing fast. We have machine learning artificial intelligence systems that are threatening our jobs and scaring people into thinking we are heading into a dystopian future. It is probably not that serious, but these machines need some new ugrades. As demand for more power, small size, and higher efficiency increases — especially in data centers, AI, EVs, and fast charging — silicon starts to really struggle.

The Real World Impact

GaN's benefits are not theoretical by any means. I would not be writing this blog post if it was just some buzzword in the market. Again, look at those phone fast chargers in the market. GaN power integrated circuits (ICs) allow those modern USB-C chargers for phones and laptops to become so much smaller than silicon ones in the past. A GaN charger can contain 60-100 W of output in a pocket-sized charging brick, where silicon ones required those abnormally large shells that come with some Dell laptops (or Xbox 360 if you remember that PSU 😮‍💨 ). The size reduction is directly related due to GaN's high-frequency switching, which shrinks bulky transformers and heat sinking requirements.

Data Centers

Data centers are feeling the impact of GaN. Enterprise and cloud data centers demand extremely efficient power conversion to supply servers, especially as power densities climb in the machine learning / AI world. Navitas, who I am super duper excited about, introduced server power supply units (PSU's) that leverage GaN to reach 98% efficiency. In comparison to the best silicon-based solution of similar power (137 W/in^3 power density).

A whole lot of numbers, but look, these GaN-based PSU's give data centers tangible benefits. With GaN, less energy is lost as heat. This means reduced cooling requirements and more power per rack without overheating. Operators can pack more compute per square foot — more GPUs, CPUs, and AI accelerators.

Silicon switches cannot turn on/off fast enough without generating a lot of heat. Due to this, it means that during each switching cycle there will be larger, slow circuits, and wasted energy which is not good for the pockets of the shareholders (or for the environment for that matter).

Switching in power electronics refers to how fast a semiconductor device is able to turn on and off. Pretty simple, right? Except the fact that this often times happens hundreds of thousands to millions of times per second.

Data centers do not just run everything at full throttle 24/7. Data centers dynamically adjust power based on demand. If a GPU suddenly ramps up, power supplies must respond instantly. If they can’t, systems throttle or crash. GaN enables faster, more stable responses.

GaN in AI Data Centers

Oooooo 👻 Artificial Intelligence. Scary, stuff. But the rise of AI has pushed data center power demands through the roof. GaN plays a big role in meeting these challenges. Compared to traditional servers, modern AI training clusters draw enormous power; for example, single AI accelerator racks can consume 100-200+ Kw per rack. Think about NVIDIA's Grace Blackwell chips — these revolutionary chips draw about 50-60x more than earlier servers.

Higher distribution voltages inside data centers are a big trend. Back in the old days, servers were powered from a 12 V DC bus, then 48 V, now 54 V DC is sometimes used. Without getting into the nitty gritty details of it all, the lower voltages lead to enormous currents for megawatt-scale racks; basically, making it impractical. So to make it practical, NVIDIA has a new initiative with Navitas: 800 V DC high-voltage distribution (HVDC) architecture for AI "factories" which incorporate GaN semiconductors.

Imagine you’re trying to send water through a long hose to fill many buckets. If the water pressure is low, you need really big hoses to get enough water through — and you lose a lot along the way (not good). But if you increase the pressure, you can use thinner hoses, send water farther, and waste less.

NVIDIA’s 800V HVDC system is like increasing the “pressure” of electricity.

• • Higher voltage means electricity flows more efficiently — with less energy lost as heat.
• • It lets data centers deliver more power using smaller, cheaper cables.
• • That’s especially important now, because the new AI chips (like Grace Blackwell) need a lot of power.

Electric Cars

Oh man, if you got one of these electric vehicles then you might be more excited about this than the data centers. Let's dive into it.

In electric vehicles (EVs), the on-board charger (OBC) is responsible for converting AC power from a wall outlet or charging station into DC power to charge the battery. In contrast, DC fast chargers bypass the OBC and deliver DC power directly to the battery using a DC-DC converter. Traditionally, silicon was used in these systems (and silicon carbide for higher voltages), but GaN, particularly for 400 V-class EV platforms, looks very compelling.

GaN-based OBCs can switch faster and with lower losses. This enables higher power levels in the same size charger. Navitas estimates that a GaN OBC can charge an EV up to 3x faster than a traditional silicon-based OBC. But wait, there is more! GaN OBCs also cut energy losses (and heat 🔥 ) by somewhere around 70%. To make this more clear, basically, when you plug your charger to the car you will get more of the grid energy into the cars battery rather than have it be wasted as heat, which speeds up your charging time and prevents your car from being destroyed by... heat.

Furthermore, if you still are not satisfied, GaN-based OBCs can could also extend an EV's range and even allow a smaller battery by around 5%. For those of you that are environmentally-concious, the efficiency could cut road CO₂ emissions by 20% per year by 2050 (yes, that aligns with climate targets... for now).

Who is Leading the Way with GaN?

Now we are thinking like a investor, which is not my forte. But as mentioned, Navitas Semiconductor. The company, started in 2014 by veterans of the power semiconductor industry, operates on a fabless model. This means it designs the integrated circuits in-house but outsources the semiconductor fabrication to foundries. While there were other companies that sold GaN transistors, Navitas released GaNFast™. Competitors required complex external drive circuitry and were not as reliable if they were not controlled properly. Navitas's GaNFast™ IC's fixed this concern by integrating gate drive, level shifting, protection features, and sensing into one package. In short, it is almost like "drop and play".

Navitas partnered with NVIDIA for its data centers, but how about in the automobile space? Changan Automobile in China announced in late 2024 that they built the first commercial GaN-based on-board charger in a production EV. This was using Navitas's GaN technology.

There are more players in this space, too. infineon, STMicro, and Onsemi are all in the GaN space. They are developing automative-grade GaN devices and modules.

So Farewell to Silicon?

Not quite. Don't get sad about it, though. I know you are a big supporter of GaN now after reading up until now, but silicon is still needed in your life.

Take for example NVIDIA 800 V architecture that is built using Navita's semiconductors. The architecture does not just use GaN, it actually also uses silicon carbide (SiC). While the rack-level DC-DC converters benefit from GaN, their SiC semiconductors provides benefits for ultra-high-voltage devices (650 V up to 6.5 kV SiC MOSFETs and diodes) for the front-end solid-state transformer and AC/DC rectification stages. By combining SiC for high voltage and GaN for medium voltage, the entire chain from grid to GPU can be optimized. If that was a whole lot to read—it means silicon is still needed.

GaN is good for power delivery, but CPUs, GPUs, and memory chips are all silicon-based logic semiconductors. Furthermore, signal processors, HDMI controllers, flight control systems, sensors, avionics are all silicon based.

Logic semiconductors are incredibly complex and very-closely tied to silicon CMOS processes. While can switch (there's that word again...) fast, logic chips need billions of tiny, low-power transistors that switch with extreme precision and complexity. Silicon CMOS are hyper-optimized with 50+ years of research and development.