A CPU looks like a small metal square from the outside, but beneath that metal lid sits a surprisingly complex sandwich of layers: a protective heat spreader on top, a thin layer of thermal material, a tiny silicon chip (the actual processor), and a circuit board substrate underneath with hundreds or thousands of electrical contacts on its bottom face. Each layer serves a distinct purpose, and the silicon chip itself contains billions of microscopic structures invisible to the naked eye.
The Layers You See When You Open One Up
If you pried the metal lid off a desktop CPU (a process enthusiasts call “delidding”), you’d find three main layers stacked together. The outermost layer is the integrated heat spreader, or IHS. It’s a nickel-plated copper cap glued to the circuit board below with epoxy. Its job is simple: spread heat from the tiny chip underneath across a larger surface area so your cooler can absorb it efficiently.
Between the heat spreader and the silicon chip sits a thin layer of thermal interface material. In high-performance processors, this is typically a thin sheet of indium or indium alloy soldered directly to the chip. Indium conducts heat extremely well, around 86 watts per meter-kelvin, which is far better than the thermal paste you might squeeze from a tube. Lower-end chips sometimes use a polymer-based thermal compound instead, which is cheaper but less effective.
Below all of that is the silicon die itself, the actual processor. It’s remarkably small, often no larger than your thumbnail. NVIDIA’s GB202 chip, one of the largest consumer-grade dies currently made, measures about 750 square millimeters, roughly the size of a postage stamp. Most desktop CPUs are considerably smaller. The die is mounted on a green or dark-colored substrate, a multi-layered circuit board that routes electrical signals from the chip to the contact points on the bottom.
What the Bottom Looks Like
Flip a CPU over and you’ll see the electrical interface, either a grid of tiny metal pins or a flat grid of gold-colored pads. The two designs are called PGA (pin grid array) and LGA (land grid array). AMD’s older desktop chips used PGA, where the processor itself has hundreds of thin gold-plated pins that slot into holes in the motherboard socket. Intel’s desktop chips and AMD’s newer ones use LGA, where the bottom of the chip has flat metal pads and the delicate spring-loaded pins live in the motherboard socket instead. LGA pads are less fragile since there are no pins to bend during installation.
These contact points aren’t just for data. Some carry power, some carry ground connections, and others handle high-speed data signals. A modern desktop CPU can have over 1,700 of these pads or pins packed into a space smaller than a cracker.
What’s Etched Into the Silicon
The silicon die is where things get truly microscopic. If you could zoom in with an electron microscope, you’d see an incredibly dense grid of transistors, the tiny switches that perform all computation. A single modern chip packs billions of them. NVIDIA’s GB202 contains 92.2 billion transistors; their data center chip, the GB100, holds 104 billion. Even a mainstream desktop CPU contains tens of billions.
Each transistor is unimaginably small. On chips labeled “5 nanometer,” the actual gate structures have a pitch (the repeating distance from one gate to the next) of about 48 nanometers. For perspective, a human hair is roughly 80,000 nanometers wide. You could fit over a thousand transistor gates side by side across the width of a single hair.
These transistors are arranged in distinct rectangular blocks visible in die photographs (often called “die shots”). The blocks aren’t random. Each one has a specific function:
- Processor cores are the rectangular blocks that do the actual math and logic. A modern desktop chip typically has 8 to 24 of these, each containing its own arithmetic units and small, fast memory caches.
- Cache memory takes up a surprisingly large portion of the die. Level 1 cache sits inside each core. Level 2 cache surrounds the cores. Level 3 cache often appears as large, uniform rectangular blocks that can occupy a third or more of the total chip area. In die shots, cache is easy to spot because it looks like a repeating grid pattern, very different from the irregular layout of a processing core.
- Memory controllers and I/O blocks sit along the edges of the die, handling communication with RAM, storage drives, and expansion slots.
- An integrated graphics processor, on chips that include one, typically occupies its own large block of the die.
In die shot photographs, these functional blocks look like a colorful aerial view of a planned city, with orderly cache “neighborhoods” next to the denser, more chaotic-looking core “downtown” areas.
Chiplet Designs: Multiple Dies in One Package
Not every CPU contains a single silicon chip. Many modern processors use a chiplet design, where multiple smaller dies sit side by side (or stacked on top of each other) inside one package. If you opened one of these, you’d see several distinct silicon chips rather than one large one.
AMD’s Ryzen desktop processors, for example, separate the processing cores onto small “compute” chiplets and put memory controllers and I/O functions on a separate “I/O” die. The compute chiplets are manufactured on a cutting-edge, expensive process, while the I/O die uses an older, cheaper one. They communicate through high-speed links built into the substrate.
This approach gets even more extreme in data center chips. AMD’s Instinct MI300 series stacks up to 8 compute dies vertically on top of 4 I/O dies, all connected together and linked to 8 stacks of high-bandwidth memory, all within a single package. NVIDIA’s B100 accelerator puts two GB100 dies in one package connected by a 10 terabyte-per-second link, totaling 208 billion transistors under one heat spreader.
How Deep the Layers Go
Even within a single silicon die, there’s hidden vertical complexity. The transistors themselves sit at the very bottom of the chip, etched into the silicon surface. Above them rise dozens of metal interconnect layers, thin copper wires that connect transistors to each other. These layers get progressively wider as they move upward, starting at nanometer-scale widths near the transistors and reaching micrometer-scale widths at the top where they connect to the substrate below.
Cross-section images taken with electron microscopes reveal this structure beautifully. The transistor layer at the bottom looks like a row of tiny identical pillars. Above it, the metal layers form an intricate lattice that resembles a multi-story parking garage, with vertical pillars (called vias) connecting each level. A modern chip can have 15 or more of these metal layers. All of this complexity exists in a slab of silicon thinner than a millimeter.
What makes this especially striking is the contrast with what you hold in your hand. From the outside, a CPU is just a plain metal square. Inside, it’s one of the most complex objects ever manufactured, with billions of switches connected by miles of microscopic wiring, all working in concert billions of times per second.