From sand to silicon. From a bit to a computer.

Silica is silicon combined with oxygen. Quartz-rich feedstock is selected and processed; a handful of ordinary sand is not ready-made chip material.

Industrial reduction changes silica into silicon. This changes the chemistry; the resulting material still needs further purification for demanding electronic uses.

Chemical processing removes unwanted impurities and produces high-purity polysilicon. Controlled doping later adds selected impurities for a purpose; it is not the same as leaving contamination behind.

A controlled crystal-growth process creates an ingot with an ordered atomic structure. Purity and crystal structure are separate properties.

An ingot is sliced and its surfaces prepared. A wafer is a starting surface for device processing, not a complete CPU or solar cell.

Different processes form active regions, insulating layers and connections. Chip making repeats many tightly controlled steps before testing, separation and packaging.

A CPU executes instructions using control logic, registers and arithmetic units. Caches and memory connections help supply its work. Modern CPUs also exploit parallelism.

Follow an instruction through fetching, decoding and execution. Register values and arithmetic results are different from data stored in an SSD.

A cache keeps selected data close to computation. A memory controller connects the processor to external memory; a larger cache is not the same thing as a larger SSD.

A GPU organizes many arithmetic workers, schedulers, registers, caches and memory connections. Parallel work helps only when the algorithm and data movement permit it.

See how groups of threads use execution lanes and shared resources. A GPU arithmetic lane and a complete CPU core are different units, so their counts are not a direct comparison.

VRAM is memory attached to a GPU. Capacity, bandwidth and latency describe different limits. Adding memory needs a supported electrical, board and software configuration.

GB202 is a GPU design used in both the RTX 5090 and RTX PRO 6000 Blackwell. The products have different enabled configurations and memory systems. Explore what is physical and what is a product choice.

Two GPU boards have separate local memories. Splitting data or a model across devices needs compatible communication paths and software; it does not turn them into one equally fast memory bank.

The same broad semiconductor material can form different ways to hold a bit. SRAM, DRAM and NAND flash trade area, speed, persistence and other properties differently.

An SRAM cell uses a feedback circuit to hold a state while powered. It needs no periodic DRAM-style refresh, but it is still volatile.

A DRAM cell represents a bit using stored charge. Charge leaks, so refresh is required. A RAM module combines memory devices and supporting connections.

Flash stores information through charge that changes a cell’s electrical behavior. 3D NAND stacks cell structures; it is different from stacking ordinary RAM sticks.

An SSD combines NAND flash with a controller, firmware and interfaces on a board. It is an assembled storage system, not one uniform silicon chip.

The controller maps logical addresses, manages errors and moves or retires physical flash blocks. NVMe describes a host interface, not a type of flash cell.

A crystalline-silicon solar cell uses light to generate electrical energy. It branches into a different device process rather than following the complete CPU fabrication route. Other photovoltaic materials also exist.

Absorbed light can create mobile charge carriers. The cell structure and contacts collect them; a connected load lets electrical energy be delivered to a circuit.

Connected cells form a module with protective layers and wiring. Series and parallel connections affect voltage and current; shading and cell mismatch complicate real systems.

Build a small Boolean circuit now, check every input and export Verilog. Then learn what synthesis, physical design and a process design kit add to the evidence.

Start with particles, charge and current. Track what moves through a wire and distinguish electron motion from the direction of conventional current.

A gate voltage can control a transistor’s conducting path. Combine complementary switches to understand an ideal CMOS inverter, then inspect where the simple switch model stops being enough.

Use powers of two to represent an unsigned number. Add two eight-bit values and follow the carry when the result exceeds the available bits.

A tiny teaching machine fetches, decodes and executes instructions. Step through a calculation and watch the program counter, accumulator and memory change.