A System on a Chip (SoC) is a single integrated circuit that holds most of the parts of an electronic system: processor cores, graphics, memory controllers, on-chip memory, connectivity and dedicated accelerators, all joined by an on-chip interconnect. Putting them on one die makes a device smaller, cheaper to build in volume and far more power-efficient than the same functions spread across separate chips. That is why the phone, watch, car dashboard and smart speaker around you each run on an SoC.
Source: Google/Unknown
Smartphones, tablets, smartwatches and smart home gadgets have become part of daily life, and at the core of each of these compact, capable devices sits an SoC. This post explains what an SoC is, what goes inside one, how the parts work together, where SoCs are used and where they are heading.
Key takeaways
- An SoC combines CPU, GPU, memory controllers, on-chip SRAM, I/O and accelerators on one die.
- The main working memory (DRAM) usually sits outside the SoC, often stacked on top of it or placed next to it in the same package.
- An on-chip interconnect (bus or network-on-chip) moves data between blocks; it is as important to performance as the CPU.
- Integration cuts board area, latency and power, which is why SoCs dominate battery-powered and embedded products.
- The trade-offs are high design cost, long schedules and less flexibility once the chip is taped out; chiplets are one response to this.
What is a System on a Chip?
A System on a Chip (SoC) is an integrated circuit that combines multiple components of an electronic system into a single chip. It typically includes a central processing unit (CPU), memory, input/output ports, and often other components such as graphics processing units (GPUs), modems, or specialised hardware for tasks like AI or signal processing. SoCs are designed to be compact, power-efficient and cost-effective, which makes them common in smartphones, tablets, IoT devices and embedded systems. The chip in your phone that handles processing, graphics and connectivity is almost certainly an SoC.
The difference from a traditional PC motherboard is the level of integration. On an older PC the processor, graphics card, chipset, network controller and audio codec were separate chips on a board, joined by copper traces. On an SoC those functions are blocks on one piece of silicon, joined by on-chip wiring that is far shorter, faster and cheaper in energy per bit.
What is inside a typical SoC?
A typical SoC includes:
- Central Processing Unit (CPU): the brain of the system, handling general-purpose computing. Mobile SoCs usually mix large high-performance cores with small efficient cores.
- Graphics Processing Unit (GPU): manages visual rendering for displays, games and user interfaces, and increasingly general parallel compute.
- On-chip memory: SRAM for caches, scratchpads and buffers that hold data and instructions close to the processors. The larger main memory (LPDDR DRAM) is normally a separate chip, reached through an on-chip memory controller.
- Modem: enables wireless connectivity such as 4G or 5G cellular. Wi-Fi and Bluetooth may be on the SoC or on a companion chip.
- Input/output interfaces: for connecting to peripherals, such as USB, PCIe, display outputs (for example HDMI or DisplayPort through a controller), camera and storage interfaces.
- Specialised hardware: blocks for specific tasks, such as AI accelerators (NPUs), image signal processors, video codecs, audio processors and security engines.
Two further pieces are easy to overlook but present in every SoC: the interconnect that links all the blocks, and the infrastructure of clocks, power management, reset, debug and test logic that keeps the chip running and makes it testable in production.
By integrating these components, SoCs strike a balance of performance, efficiency and compactness, which makes them ideal for portable and embedded devices.
How does an SoC work?
An SoC acts as the central hub of a device, coordinating and executing tasks. Each block communicates through an internal bus or interconnect, so data can move between them without leaving the chip. For example:
- The CPU runs the operating system and application code.
- The GPU renders graphics for video, games and the user interface.
- The modem connects the device to cellular networks and the internet.
- Specialised hardware accelerates tasks such as machine learning or image processing.
Keeping these components on one chip reduces latency, because data does not have to travel between separate packages, and improves power efficiency, which is critical for battery-powered devices.
The interconnect: buses and networks-on-chip
Small microcontroller-class SoCs often use a simple shared bus, such as Arm AMBA AHB or APB. Larger SoCs need many blocks to talk at once, so they use a crossbar or a network-on-chip (NoC), usually with AMBA AXI or a similar protocol at each block's edge. Cache-coherent interconnects let the CPU, GPU and accelerators share data without software copying it back and forth. Interconnect design decides much of the real-world performance: a fast CPU starved of memory bandwidth will not deliver its headline numbers.
Power management
An SoC is divided into power domains and clock domains. Blocks that are idle are clock-gated or switched off completely, and active blocks run at a voltage and frequency chosen for the current load (dynamic voltage and frequency scaling, DVFS). Because dynamic power scales roughly with C x V² x f, lowering supply voltage gives the largest saving.
A worked example: why integration saves energy
The numbers below are illustrative, chosen only to show the order of magnitude. Suppose moving one bit costs about 0.1 pJ across a few millimetres of on-chip wire, and about 5 to 10 pJ when it has to leave the chip and cross a board to another package. A camera pipeline moving 2 GB/s (16 Gbit/s) between an image processor and an AI block would then spend:
| Path | Energy per bit (illustrative) | Power at 16 Gbit/s |
|---|---|---|
| On-chip, between two SoC blocks | 0.1 pJ | about 1.6 mW |
| Off-chip, between two packages on a board | 5 to 10 pJ | about 80 to 160 mW |
A difference of tens to over a hundred milliwatts on one data stream matters a great deal in a phone with a total sustained budget of a few watts. Multiply that across camera, display, memory and modem traffic and the case for integration is clear.
Why are SoCs important?
SoCs have changed the electronics industry for several reasons:
- Size: combining many functions on one chip enables smaller, slimmer devices without giving up performance.
- Power: integrated designs consume less energy, extending battery life in portable devices.
- Cost: in high volume, one chip is often cheaper than several specialised chips plus the board area, assembly and testing they need.
- Optimisation: SoCs are tailored to their target devices, so designers can tune them for gaming, AI, camera quality or connectivity.
The trade-offs
- Design cost and time: a leading-edge SoC needs large teams, licensed IP, EDA tools and expensive mask sets, so it only makes sense at high volume.
- Less flexibility: a bug or missing feature means a new silicon spin, which takes months.
- Yield: a very large die is more likely to contain a defect, which is one reason some designs are moving to chiplets.
- Thermal density: packing CPU, GPU and NPU close together concentrates heat, so sustained performance is often limited by cooling.
SoC versus microcontroller versus chiplets
| Approach | What it is | Typical use |
|---|---|---|
| Microcontroller (MCU) | A small SoC with CPU, flash, SRAM and peripherals on one chip | Appliances, sensors, motor control |
| Application SoC | Multi-core CPU, GPU, NPU, modem and controllers, with external DRAM | Phones, tablets, laptops, infotainment |
| Chiplet-based system | Several dies, each built on the best process for its job, joined in one advanced package | Servers, AI accelerators, some PC processors |
Where are SoCs used?
SoCs are found throughout modern electronics:
- Smartphones and tablets: chips such as Apple's A-series and Qualcomm's Snapdragon power most mobile devices.
- Wearables: smartwatches and fitness trackers rely on SoCs for processing and connectivity in tiny form factors.
- Smart home and IoT: thermostats, security cameras and other connected devices use SoCs for efficient operation.
- Automotive and medical: SoCs power everything from car infotainment and driver assistance to medical devices.
- Consoles and PCs: some game consoles and low-power PCs use SoCs for integrated performance.
Some well-known SoCs
- Apple M-series (for example M1, M2) used in MacBooks and iPads, known for high performance and efficiency.
- Qualcomm Snapdragon powers many Android smartphones, with strong CPU, GPU and modem capabilities.
- MediaTek Dimensity found in many mid-range and flagship smartphones, balancing cost and performance.
- Broadcom SoCs in the Raspberry Pi used in single-board computers for hobbyists and developers.
How an SoC is built
Few SoC teams design every block themselves. A typical project licenses CPU cores, interconnect and interface IP (USB, PCIe, LPDDR PHYs), designs its own differentiating blocks, and then integrates and verifies the whole. The main steps are architecture and specification, RTL design, verification (simulation, emulation and formal methods), synthesis, physical design, timing and power sign-off, tape-out, and then silicon bring-up and production test. Verification usually takes the largest share of the effort.
Common misunderstandings
- "All the memory is on the chip." Caches and buffers are, but the gigabytes of main memory are normally separate DRAM chips.
- "More cores means a faster SoC." Memory bandwidth, interconnect, thermal limits and software decide how much of that compute is usable.
- "SoC means mobile." SoCs are just as common in cars, networking equipment, TVs and industrial control.
The future of SoCs
As technology evolves, so do SoCs. Trends to watch include:
- AI integration: more SoCs include dedicated AI accelerators for voice recognition, image processing and on-device language models.
- 5G and beyond: advanced modems enable faster, more reliable connectivity.
- Energy efficiency: newer manufacturing processes, such as 3 nm and 2 nm class nodes, make SoCs more power-efficient.
- Customisation: manufacturers are designing SoCs tailored to specific applications, such as automotive or augmented reality.
- Chiplets and advanced packaging: some functions are moving onto separate dies in one package, so each part can use the most suitable process.
FAQ
Is a CPU the same as an SoC?
No. A CPU is one block that executes instructions. An SoC contains one or more CPUs plus many other blocks, such as graphics, memory controllers, I/O and accelerators.
Does an SoC include RAM?
It includes SRAM caches and buffers. The main DRAM is usually a separate chip, often mounted on top of the SoC (package-on-package) or alongside it in the same package.
Why do phones use SoCs instead of separate chips?
Space, battery life and cost. An SoC needs less board area and moves data on-chip at a fraction of the energy needed to send it between packages.
What is the difference between an SoC and a chiplet design?
An SoC puts the whole system on one die. A chiplet design splits it across several dies joined in one package, which can improve yield and lets each die use a different process.
Your comments will be moderated before it appears here.