70 terms / One connected reference
From integrated-circuit fundamentals to AI processors, stacked memory, and robotics. Find a term, see it in context, and follow the connections.
70 terms · All topics
Every term has a full page: click a term name to explore its explanation, examples and related concepts. The # links still point to entries in this directory. Start with integrated-circuit basics or explore our four AI & robotics guides. For additional terminology, visit Hybrid’s full electronics glossary.
2.5D packaging commonly connects dies arranged side by side through a high-density interposer or related interconnect structure within one package. (Samsung: heterogeneous integration)
In context. An accelerator package can place logic next to HBM stacks. The package can be described as 2.5D even though the memory itself contains vertically stacked dies.
3D stacking places semiconductor dies above one another with connections between layers. Different technologies use different bonding and vertical interconnect methods. (Samsung: heterogeneous integration)
In context. “3D” describes an integration arrangement, not a universal interface. Consult the package description to understand which dies are stacked and how they connect.
An actuator converts energy into mechanical work or force. Robots use actuators such as electric motors, pneumatic devices, and hydraulic mechanisms to move. (University of Rochester: actuators)
In context. An actuator is not necessarily an IC. A robot’s electronics sense, compute, and drive; the actuator is the physical mechanism that performs the movement.
An analog-to-digital converter changes a continuously varying electrical input into digital values that a digital system can process. (Hybrid Electronics glossary)
In context. A sensor may produce a voltage that changes with temperature. An ADC lets a digital system work with measurements of that voltage, rather than the original analog signal.
An AI accelerator is hardware, or a hardware feature, optimized to perform operations used in AI workloads. It may be a dedicated device or a processing block integrated with other functions. (Intel: AI accelerators)
In context. A GPU, NPU, or purpose-built accelerator can play this role. Read the supported workloads and software requirements rather than assuming every accelerator runs the same models.
An analog IC works with continuously varying signals, such as voltages or currents, rather than only discrete logic states.
In context. Think of an amplifier that makes a small signal larger. “Analog” describes the signal-processing function, not the size or shape of the package.
ASIC stands for application-specific integrated circuit: an IC designed for a particular application rather than a general-purpose role. (Hybrid Electronics glossary)
In context. The abbreviation describes a chip’s intended role. It does not, by itself, tell you its package, supply voltage, interfaces, or whether it can be substituted for another device.
A BGA package uses an array of solder balls on its underside to make connections to a circuit board.
In context. A BGA and a leaded package can contain devices with similar functions while requiring entirely different board footprints. Check the package drawing, not just the family name.
CAN is a message-based serial communications bus that allows multiple nodes to communicate. Common high-speed CAN physical layers use differential signaling on a pair of wires. (Texas Instruments: CAN fundamentals)
In context. Separate the CAN controller, transceiver, and higher-level application protocol when reading a design. Do not apply the data-rate limits of one CAN variant to every CAN implementation.
A capacitor stores electrical energy using conductors separated by an insulating material called a dielectric. (Hybrid Electronics glossary)
In context. Capacitors appear in filtering and signal-coupling circuits. When reading an IC application circuit, distinguish capacitors built into the IC from the separate capacitors shown around it.
A chiplet is a semiconductor die designed to operate alongside other dies as part of a larger system. Multiple chiplets can be assembled into one packaged component. (Arm: chiplets)
In context. A design may separate compute, memory, and input/output functions across dies. Chiplets are not automatically interchangeable parts that can be mixed without interface and package engineering.
CMOS is an integrated-circuit technology that uses complementary p-channel and n-channel MOS devices. (Hybrid Electronics glossary)
In context. “CMOS” identifies a technology, not a universal compatibility standard. Two CMOS devices should not be treated as interchangeable without comparing their electrical specifications.
A CPU executes program instructions and handles general-purpose computation and operating-system work. Its flexible processing cores support a wide variety of tasks. (Intel: CPU vs. GPU)
In context. When reading a system diagram, look for the CPU’s coordinating role as well as any GPU or NPU blocks. “CPU” is a function, not a guarantee that all processing lives on a separate chip.
A depth camera produces an image or data map whose values describe distance from the camera. Some devices provide both depth and color information, often called RGB-D. (RealSense: depth cameras)
In context. Stereo, structured-light, and time-of-flight systems measure depth in different ways. Select the technology for the actual scene and lighting rather than treating all depth cameras as identical.
A die is the small piece of semiconductor material containing a fabricated circuit before it is considered together with its external package.
In context. A useful distinction is die versus package: the die contains the circuit; the package protects it and provides connections that are easier to use on a board.
A die-to-die interconnect carries signals between semiconductor dies within a package. The physical connection and communication protocol are both part of making those dies work together. (ASE: UCIe specification)
In context. Do not confuse an internal die-to-die link with a board-level connector. Check the interface specification and packaging requirements for the particular design.
A digital IC processes discrete logic states, commonly represented as binary 0 and 1, to perform operations such as logic or data storage.
In context. A logic gate, a processor, and a memory device serve different purposes, even though all can be digital ICs. “Digital” is a broad category, not a specific function.
A DIP has two parallel rows of pins and is a familiar through-hole package for integrated circuits.
In context. DIP packages are common in breadboard demonstrations. The pin count alone is not enough to select a replacement: verify the body dimensions, pinout, and electrical function too.
DRAM stores data as charge in memory cells that require periodic refresh. It is volatile: the stored information is lost when power is removed. (Samsung: DRAM and SRAM)
In context. HBM is a DRAM architecture, not a replacement term for all memory. Identify the interface and organization as well as the capacity when comparing memory devices.
DSP refers to processing signals represented digitally. In hardware descriptions, it can also refer to a processor optimized for that work; DSP instructions may instead be built into a CPU. (Arm: digital signal processing)
In context. Audio, voice, and sensor applications are useful places to encounter DSP. Check whether the document means a signal-processing technique, a dedicated processor, or instruction-set extensions.
Edge AI runs AI applications near the devices or environments producing the data, rather than relying entirely on a centralized cloud or data center. (NVIDIA: edge AI)
In context. In a robot example, a model can analyze sensor data locally. “Edge” describes where computation happens; it does not identify one chip type or mean the system never connects to the cloud.
An encoder provides position or movement information that a control system can use as feedback. Motor-control systems can use incremental or absolute encoders. (Texas Instruments: motor control)
In context. Wheel encoder counts can feed odometry, while a joint encoder can support motion control. Review how a particular encoder represents position and connects to its controller.
EtherCAT is an industrial Ethernet technology designed for real-time automation communication. Devices process data as frames pass through the network. (EtherCAT Technology Group)
In context. In a multi-axis machine, communication timing and synchronization matter alongside processor speed. A normal Ethernet connector alone does not establish EtherCAT device capability.
An FPGA is a type of programmable integrated circuit used to implement digital logic.
In context. Reading “FPGA” tells you the broad device class. For an actual design, use the manufacturer’s documentation to determine the available logic resources, interfaces, package, and configuration requirements.
A GPU is a processor designed to perform many operations in parallel. Originally associated with graphics, programmable GPUs also execute suitable AI and other compute workloads. (Intel: CPU vs. GPU)
In context. A GPU can render an image or process parallel calculations for a neural network. That does not mean it replaces the CPU or accelerates every kind of program.
HBM is a stacked DRAM architecture that uses a wide interface to transfer large amounts of data. Multiple memory dies are connected vertically and can be packaged close to a processor. (Micron: high-bandwidth memory)
In context. On an accelerator diagram, distinguish the compute die from the HBM stacks supplying data. HBM is memory, not the processor performing the AI calculations.
Heterogeneous integration combines components with different functions, technologies, or manufacturing processes into a larger integrated system, often within one package. (Samsung: heterogeneous integration)
In context. A package combining logic and memory is a useful example. The term describes the integration approach, not a guarantee of lower cost or higher performance.
An IMU combines inertial sensors, typically accelerometers and gyroscopes, to measure acceleration and angular motion. Its readings are inputs to motion-estimation algorithms. (Analog Devices: IMU documentation)
In context. Do not treat raw IMU output as an automatically accurate position. Calibration and the estimation method matter when combining it with camera or other sensor data.
Inference is the use of a trained AI model to produce an output from new input data. It is distinct from the training process that creates or updates the model. (NVIDIA: edge AI)
In context. Recognizing an object in a new camera image is an inference task. Learning the model from examples is training; evaluating a deployed model still requires application-specific testing.
An integrated circuit combines interconnected electronic components on a semiconductor chip to perform an electronic function.
In context. A small amplifier IC and a complex microprocessor are both integrated circuits. The phrase does not imply a particular pin count, package, or level of complexity.
An interposer provides routing and connections between dies and the surrounding package structure. Advanced packages can use different interposer materials and constructions. (Samsung: heterogeneous integration)
In context. In a 2.5D example, logic and HBM can sit side by side on a silicon interposer. Not every multi-die package uses that exact arrangement.
LiDAR measures distance using light, commonly laser pulses, and can produce spatial measurements organized as point clouds. Those measurements can support robotic mapping and localization. (MathWorks: SLAM)
In context. The LiDAR sensor supplies measurements; a SLAM algorithm uses measurements to estimate a map and location. They are related but not the same thing.
A logic gate performs a logic operation on its inputs to produce an output, and gates can be combined into larger digital circuits.
In context. A device can contain a few individual gates or combine many gates into a more complicated function. Start with the function diagram when learning what a logic IC does.
In SLAM, loop closure identifies a return to a previously visited location so the system can correct accumulated mapping and localization error. (MathWorks: SLAM)
In context. A robot that recognizes a familiar corridor can use that match to improve its map. The phrase describes a mapping operation, not a motor-control feedback loop.
Machine vision uses cameras and processing to inspect, identify, measure, or locate objects so an automated system can make decisions. (Cognex: machine vision)
In context. A robot locating a component before picking it is a useful example. Machine vision is a system capability, not the name of one camera chip or AI model.
Memory is the part of an electronic system used to receive and store information for later use. (Hybrid Electronics glossary)
In context. For a specific memory IC, identify the memory type before comparing parts. The word “memory” alone does not establish how data is accessed or whether it survives removal of power.
Memory bandwidth describes the rate at which data can be transferred through a memory interface, commonly expressed in gigabytes or terabytes per second. (Micron: high-bandwidth memory)
In context. Keep storage capacity and transfer rate separate: GB describes an amount of data, while GB/s describes a rate. A peak interface specification is not a promise of sustained application performance.
A microcontroller is an integrated circuit used as a programmable control device in an electronic system.
In context. For a control application, compare the device’s documented processing, memory, and input/output capabilities. The package and pin count are only part of the selection.
A microprocessor implements the functions of a computer processor in a chip or a small group of chips. (Hybrid Electronics glossary)
In context. Do not assume that “processor” means a complete computer on its own. Check which memory, power, clock, and peripheral functions the actual device needs from the rest of the system.
A mixed-signal IC combines analog and digital functions within the same integrated circuit.
In context. A sensor interface may need to handle a small analog signal and deliver digital data. Calling it mixed-signal helps describe that combination, rather than placing it in only one category.
A MOSFET is a transistor in which an electric field associated with the gate controls conduction between the source and drain. (Hybrid Electronics glossary)
In context. Keep the device and circuit levels separate when reading a diagram: a transistor is a building block, while an IC can combine many interconnected devices into a larger function.
A motor driver provides the electrical drive needed to operate a motor under control commands. Products differ in how much control logic and power circuitry they integrate. (Texas Instruments: motor control)
In context. Distinguish the controller deciding what motion is wanted from the driver delivering electrical drive. Some products combine these roles; check the block diagram.
An NPU is a specialized processor or processing block designed to accelerate neural-network computations. NPUs are one type of AI accelerator. (Intel: AI accelerators)
In context. For a device running image or speech models, ask which model operations its NPU supports. The NPU label alone does not establish model compatibility or actual application speed.
Odometry estimates changes in position and orientation from motion-related measurements. Wheel odometry, for example, uses encoder measurements and a model of the vehicle. (MathWorks: wheel odometry)
In context. For a two-wheel differential-drive example, left and right encoder counts contribute to estimates of position and heading. An estimate is not an independent measurement of absolute location.
An operational amplifier is an amplifier commonly available as an integrated circuit for analog signal-processing applications.
In context. Look at the complete circuit, not just the IC symbol. The surrounding connections and components are important to understanding what a particular amplifier circuit is intended to do.
An IC package protects the die and provides external terminals that connect its circuit to other components.
In context. Function and package are separate questions. “Voltage regulator” describes a function; “SOIC” describes a package family. You need both electrical and mechanical details to identify a usable part.
A printed circuit board is an insulating base carrying the conductive paths used to connect electronic components. (Hybrid Electronics glossary)
In context. The board and the IC are not the same thing. An IC is one component that may be mounted on a PCB alongside resistors, capacitors, connectors, and other devices.
A pinout identifies the location and function of a device’s external connections, including power, ground, inputs, and outputs.
In context. Use the manufacturer’s drawing to establish viewing direction and pin 1. A package that physically fits a board is not necessarily electrically compatible with it.
A QFN is a surface-mount package with connection pads around the underside edges rather than long projecting leads.
In context. Do not select a board footprint from the letters QFN alone. Compare the manufacturer’s dimensions, pad arrangement, pin count, and any exposed-pad requirements.
A QFP is a surface-mount package with leads extending from all four sides of its body.
In context. QFP and QFN sound similar but make their board connections differently. Always use the exact package designation and drawing rather than relying on a familiar-looking abbreviation.
In AI, quantization reduces the numerical precision used to represent a model. It can reduce model storage and computation requirements, but the effect on accuracy must be evaluated. (NVIDIA: quantization)
In context. Do not compare two accelerator claims without checking their numerical formats. A lower-precision deployment should be tested with the actual model and hardware rather than assumed equivalent.
Real-time control requires responses within specified timing constraints, not simply high average computing speed. Embedded operating systems can help schedule work to meet those constraints. (FreeRTOS: RTOS fundamentals)
In context. For a motor-control example, define the deadline for reading feedback and updating the drive. Adding a faster processor does not by itself demonstrate that every deadline will be met.
ROS 2 is a collection of software libraries and tools for developing robot applications. It supports the software side of robotics rather than describing a semiconductor device. (ROS 2 documentation)
In context. Keep the layers separate: ROS 2 helps build robot applications, an operating system manages computing resources, and ICs execute the software and interface with hardware.
An RTOS is an operating system designed for predictable response to events under timing constraints. Its scheduler manages tasks, often using priorities. (FreeRTOS: RTOS fundamentals)
In context. An RTOS is software, not a chip. Task priorities help organize execution, but they cannot make impossible timing requirements achievable.
Semiconductor material, typically silicon in the ICs described here, forms the substrate on which integrated-circuit components are fabricated.
In context. “Semiconductor” and “integrated circuit” describe different things. The first can describe a material or a broad device category; the second describes an interconnected electronic circuit.
Sensor fusion combines measurements from different sensors to estimate a system’s state or surroundings. In robotics, it can combine cameras, IMUs, LiDAR, and other inputs. (MathWorks: SLAM)
In context. A navigation system might combine inertial motion readings with visual observations. Calibration and consistent interpretation of the measurements remain important.
A servo controller uses feedback to control a motor or actuator toward a requested state, such as a position or velocity. It is part of a closed-loop motion system. (University of Rochester: actuators)
In context. A joint may compare its target angle with measured position, then adjust its drive. “Servo” describes feedback control, not one universal package or interface.
SLAM is a computational method that builds a map of an environment while estimating the robot’s or vehicle’s location within that map. (MathWorks: SLAM)
In context. SLAM is software and mathematics, not a type of integrated circuit. Its sensor processing and optimization run on computing hardware.
An SoC integrates processing and other system functions, such as memory subsystems, interfaces, and accelerators, into an integrated circuit. (Arm: SoC development)
In context. A block diagram may show a CPU, GPU, and NPU within one SoC. Integration does not mean that the finished system needs no external memory, power circuitry, or other components.
An SOIC is a small surface-mount package with leads along two sides, often compared with a smaller, surface-mount version of a DIP.
In context. The same general package name can cover different body widths and pin counts. A specific manufacturer package drawing is a better guide than a photograph.
SRAM stores data using transistor-based cells that do not need periodic refresh while powered. “Static” does not mean nonvolatile; power is still needed to retain its contents. (Samsung: DRAM and SRAM)
In context. When learning memory terminology, keep SRAM’s lack of refresh separate from long-term storage. Compare its role in a device with the separate DRAM memory subsystem.
Surface-mount components are attached to conductive pads on a circuit board’s surface rather than using leads inserted through the board.
In context. SOIC, QFP, QFN, and BGA are different surface-mount package styles. The mounting method alone does not tell you the device’s function or assembly requirements.
Through-hole components have leads that pass through holes in a circuit board and are connected by soldering.
In context. A DIP is a familiar through-hole IC package. When converting a prototype into a board design, keep the package choice separate from the circuit’s electrical requirements.
A TPU is a Google-developed application-specific integrated circuit designed to accelerate machine-learning workloads, especially large matrix operations. (Google Cloud: TPU introduction)
In context. TPU is not a generic name for every AI chip. Check the specific TPU generation, supported software, and workload requirements when reading a platform description.
Training is the process of developing an AI model from data so that it can perform a task. A trained model can then be deployed to make predictions through inference. (NVIDIA: edge AI)
In context. Separate the computer used to train a model from the hardware that runs it in a robot. They need not be the same device or even be in the same location.
Transistors are electronic devices that can be integrated and interconnected with other components to create the functions inside an IC.
In context. It helps to think in levels: transistors are devices, an IC connects devices into a circuit, and a circuit board connects ICs and other components into a system.
A TSV is an electrical connection passing through silicon. TSVs are used in structures such as stacked HBM dies to carry signals vertically. (Micron: high-bandwidth memory)
In context. A TSV is not the same as a plated hole in a circuit board. It belongs to the semiconductor or packaging structure rather than the ordinary PCB interconnect.
UCIe is an open specification for communication between chiplets within a package. Its scope includes physical interconnect and protocol aspects of die-to-die communication. (ASE: UCIe specification)
In context. Treat UCIe support as one part of compatibility, not permission to combine arbitrary chiplets. Review the relevant version, implementation, and system requirements.
A voltage regulator is a circuit used to control a supply voltage, and voltage-regulator functions are commonly available as ICs.
In context. For a real application, start with the manufacturer’s operating conditions and recommended circuit. Do not infer the allowable input voltage, current, or heat dissipation from the package size.
A semiconductor wafer is used in fabrication to form the layers and components that will become integrated-circuit dies.
In context. Keep the manufacturing sequence clear: wafer processing forms circuits, individual dies are separated, and packaging makes the circuits practical to connect and handle.
Try a shorter phrase, an abbreviation such as “IC”, or a different topic.
Definitions are introductory, not component specifications. Use the manufacturer’s datasheet for operating limits, pinouts, footprints, and design decisions. About this resource
70 standalone definitions
Every term in the A–Z directory now has a full page. Start with these building blocks, then follow related definitions into computing and robotics.
An integrated circuit, or IC, combines interconnected electronic devices on semiconductor material to perform one or more electronic functions.
Read the Integrated circuit definition →Glossary / SemiconductorA semiconductor is a material whose electrical conductivity depends on its structure, mobile charge carriers and conditions such as temperature.
Read the Semiconductor definition →Glossary / TransistorA transistor is a semiconductor device that controls current and can be used to switch or amplify electrical signals.
Read the Transistor definition →Glossary / CPUA CPU, or central processing unit, executes program instructions and performs general-purpose processing tasks in a computing system.
Read the CPU definition →Glossary / MicrocontrollerA microcontroller, or MCU, combines a CPU, memory and input/output resources in an integrated circuit used to control an embedded application.
Read the Microcontroller definition →Glossary / System-on-chipA system-on-chip, or SoC, integrates processing and other system functions, such as memory subsystems, interfaces and accelerators, into an integrated circuit.
Read the System-on-chip definition →Glossary / GPUA GPU, or graphics processing unit, is a processor designed to perform many suitable calculations in parallel, supporting graphics rendering and other compute-intensive workloads.
Read the GPU definition →Glossary / NPUAn NPU, or neural processing unit, is a specialized hardware accelerator designed to execute neural-network operations.
Read the NPU definition →Glossary / HBMHBM, or high-bandwidth memory, is a stacked DRAM architecture that uses a wide interface to move large amounts of data between memory and a processing device.
Read the HBM definition →Glossary / ChipletsA chiplet is a semiconductor die designed to work as part of a larger system, with multiple dies combined into a packaged component.
Read the Chiplets definition →