SoC

On this page 7

An SoC (system on a chip) is an integrated circuit that combines most or all of a computer’s main functions, such as CPU cores, a GPU, memory controllers, input/output interfaces and specialized accelerators, on a single die or tightly integrated package. In semiconductors, it represents the highest level of functional integration, replacing what once required a motherboard full of separate chips.

SoCs power smartphones, tablets, many laptops, televisions, cars, routers and embedded controllers. Integration shortens the distance signals travel, which cuts power consumption and latency, reduces board area and lowers cost at high volume. The price of that integration is less flexibility after manufacture and a very expensive design process.

How an SoC integrates a complete system

An SoC integrates a system by placing functional blocks, called IP blocks, on one piece of silicon and connecting them through an on-chip interconnect or network-on-chip. Each block, whether a CPU cluster, a video decoder or a USB controller, is designed once and reused across products, often licensed from specialist IP vendors.

The interconnect manages shared access to memory, which is the central resource. A system-level cache and a unified memory controller let the CPU, GPU and neural processing unit work on the same data without copying it between separate memory pools. The table lists typical blocks found in a modern application SoC.

BlockFunctionWhy it is integrated
CPU clusterGeneral-purpose processingCore of the system, close to cache
GPUGraphics and parallel computeShares memory with the CPU
NPUNeural network inferenceHigh efficiency for AI workloads
ISPCamera image processingHandles high data rates from sensors
Media enginesVideo encode and decodeFar lower power than CPU decoding
Memory controllerInterface to LPDDR or DDR memoryLow latency, high bandwidth
Modem and connectivityCellular, Wi-Fi, BluetoothSaves board area in phones
Security enclaveKey storage, secure bootIsolated root of trust
Typical functional blocks of an application SoC

Monolithic, chiplet and package-level SoC designs

SoCs are built in three broad physical forms. A monolithic SoC places every block on one die, which gives the best power efficiency but becomes costly as dies grow, because defect probability rises with area. A chiplet-based design splits functions across several smaller dies linked inside one package. A system in package (SiP) stacks or places separate chips, often including memory, side by side.

ApproachStrengthLimitation
Monolithic SoCLowest power and latency between blocksYield falls as die size grows
Chiplet packageMix process nodes, better yieldDie-to-die links cost power
System in packageCombines different chip types quicklyLarger and less integrated
Package-on-package memoryMemory stacked above the SoCThermal limits from stacking
SoC construction approaches compared

Chiplets have blurred the boundary: a large laptop or server processor may be marketed as an SoC while being assembled from several dies on different process nodes, with I/O on an older, cheaper node and compute cores on the most advanced one. The defining feature remains that the package delivers a complete system to the board.

SoC compared with CPUs, ASICs and FPGAs

An SoC differs from neighboring chip categories in scope and programmability. A standalone CPU needs an external chipset and graphics; an ASIC is a chip built for one task; an FPGA is reconfigurable after manufacture. Many SoCs are themselves ASICs in the strict sense, because they are custom designs, while some FPGAs integrate CPU cores and are sold as SoC FPGAs.

CategoryScopeFlexibilityTypical use
SoCComplete computing systemSoftware-programmable CPU and GPUPhones, laptops, embedded devices
Discrete CPUProcessor cores and cacheGeneral purposeDesktops and servers with chipsets
ASICOne fixed functionNone after fabricationNetworking, crypto, AI accelerators
FPGAReconfigurable logicHardware reprogrammablePrototyping, low-volume products
MicrocontrollerSmall CPU with memory and peripheralsFirmware-programmableAppliances, sensors
SoC and related semiconductor categories

A microcontroller is technically a small SoC too, but the term SoC usually implies application-class processors running a full operating system, external DRAM and significant graphics or media capability.

Design trade-offs and selection criteria

Designing an SoC means balancing performance, power, area and cost, a trade-off engineers abbreviate as PPA. Every block added consumes die area and leakage power even when idle, so designers rely on power gating and dynamic voltage and frequency scaling to shut down unused regions and slow others.

For product teams choosing an existing SoC rather than designing one, the decision depends on workload, lifetime and ecosystem. The table summarizes typical choices.

Product needKey SoC propertyReason
Battery-powered handheldEfficient cores, strong media enginesLong runtime under video and browsing
Industrial controllerLong availability, wide temperature rangeProducts ship for ten years or more
Smart cameraCapable ISP and NPUOn-device vision without cloud
Automotive cockpitFunctional safety certificationRequired by ISO 26262 processes
Thin laptopUnified memory, integrated GPUPerformance without a discrete GPU
Choosing an SoC for a product

Limitations and common misconceptions

The main limitation of an SoC is that components cannot be upgraded independently. Memory is often packaged with the chip, so capacity is fixed at purchase, and a weak GPU cannot be swapped out. Thermal design is also harder, because all heat concentrates in one small package.

  • An SoC is not one specific architecture: it can use Arm, x86, RISC-V or other instruction sets.
  • Core count alone says little; cache, memory bandwidth and accelerators often dominate real performance.
  • Design costs are enormous at leading nodes, which favors high-volume products.
  • Software support decides longevity: drivers for proprietary blocks may stop receiving updates.
  • Integration does not eliminate the board; power delivery, memory and connectors still matter.

The short version

An SoC is like an entire computer shrunk onto one chip. Instead of separate parts for thinking, drawing graphics, playing video and talking to networks, everything sits together on a single piece of silicon.

Keeping all parts close together saves energy and space, which is why phones can be so thin and still last a whole day. The trade-off is that you cannot replace one part on its own: if you need more memory or graphics power later, you usually need a new device.

Where it makes a difference

A company building a battery-powered handheld barcode scanner for warehouses compares two designs. The first uses a standalone processor, separate Wi-Fi module, external graphics controller and camera interface chip. The second uses a single application SoC with integrated Wi-Fi, ISP and a small NPU, plus LPDDR memory stacked on the package.

The SoC design reduces the main board by roughly half its area, removes several power rails and lets the camera image be decoded on the NPU instead of the CPU. In field tests, a full shift of scanning finishes with battery to spare, while the multi-chip prototype needed a midday swap. The engineers also choose an SoC variant with a ten-year availability commitment, because warehouse equipment stays in service far longer than consumer phones. The outcome is a smaller, longer-running device at a lower unit cost once production volume is reached.