What Is an SoC? The Parts of a System on Chip
A system on chip (SoC) puts what used to be a whole board onto one piece of silicon: one or more processor cores, on-chip memory, a controller for external DRAM, peripherals such as UART, SPI, I2C, timers and GPIO, and often accelerators for graphics, signal processing or machine learning. An on-chip interconnect, usually built on the Arm AMBA protocols such as AXI and APB, lets every part reach the others by address. The phone in your pocket, a Wi-Fi router and an FPGA board with a Zynq device all run on SoCs.
What you'll learn
- The main blocks of an SoC and what each does
- Why everything is reached through a memory map
- Masters and slaves on the interconnect
- Why integrating on one die saves power and cost
- Where FPGAs fit: soft and hard SoCs
The blocks
| Block | What it does | Typical bus |
|---|---|---|
| CPU cores | Run the software; often with caches and an MMU | AXI master |
| On-chip SRAM | Fast scratch memory and boot code | AXI slave |
| DRAM controller | Turns bus requests into DDR commands for external memory | AXI slave |
| Peripherals | UART, SPI, I2C, GPIO, timers | APB slave, behind a bridge |
| DMA engine | Copies data between memory and peripherals without the CPU | AXI master and slave |
| Accelerators | Custom hardware for one job, such as a video codec or neural network | AXI master and slave |
Everything has an address
Software reaches every block through the memory map. A store to one address writes on-chip SRAM, to another a UART's transmit register, to another a control register of the accelerator. The interconnect decodes the address and routes the transfer to the right slave. That is why adding a peripheral to an SoC is mostly a matter of giving it an address range and a bus interface, which the AXI4-Lite project in this series does.
Masters and slaves
A master starts transfers: CPU cores, DMA engines and accelerators that fetch their own data. A slave only answers: memories and peripheral registers. Some blocks are both, such as an accelerator with control registers (slave) that also reads its input from DRAM (master). The interconnect arbitrates when two masters want the same slave.
Why put it all on one chip
- Power: driving a signal across a board takes far more energy than across a die.
- Speed: on-chip wires are short, so buses can be wide and fast.
- Cost and size: one package instead of many, and a smaller board.
- The trade-off is design effort: an SoC is expensive to design and verify, so it pays off only in volume or when it is built on an FPGA.
Soft and hard SoCs on FPGAs
On an FPGA you can build an SoC two ways. A soft SoC builds the processor itself from FPGA logic, such as AMD's MicroBlaze. A hard SoC has the processor already in silicon beside the FPGA fabric, such as the dual Arm Cortex-A9 in a Zynq 7000. Both use AXI to connect your own logic, which is why FPGA SoC projects are a good way to learn SoC design hands on.
Common mistakes
- Thinking an SoC is just a CPU. Most of the area is often memory, interconnect and peripherals.
- Ignoring the memory map until late, then finding address clashes.
- Forgetting that a block can be both master and slave.
- Assuming on-chip means instant: interconnect arbitration and DRAM latency still dominate many workloads.
Check your understanding
- Q1. What decides which block a CPU store reaches? Answer: the address, decoded by the interconnect using the memory map.
- Q2. Name a block that is both a master and a slave. Answer: a DMA engine or an accelerator with its own control registers.
- Q3. What is the difference between MicroBlaze and the Cortex-A9 in a Zynq? Answer: MicroBlaze is a soft core built from FPGA logic; the Cortex-A9 is hard silicon beside the fabric.
The SoC series
Related
Sources
Last verified
Checked against the AMBA AXI specification and the Zynq 7000 technical reference manual on 17 November 2026.