What is a logic chip? How is it different from a memory chip?

Semiconductors can be classified according to their structure into: integrated circuits, discrete devices, optoelectronic devices, and sensors. Integrated circuits are what we usually call chips, which can be divided into: digital chips and analog chips. Digital chips include: logic chips, memory chips, MCUs, etc.

What is a logic chip?

According to the World Semiconductor Trade Statistics (WSTS), all semiconductors are categorized based on their structural functions into four main types: integrated circuits, discrete devices, optoelectronic devices, and sensors.

Integrated Circuits, abbreviated as IC, are micro-structures that integrate the necessary transistors, resistors, capacitors, and other components of a logic circuit onto a semiconductor wafer using specific processes to perform desired digital circuit functions. They account for 83% of the global semiconductor market share.

Integrated circuits can be further divided into logic chips that handle computing functions, memory chips for digital data storage functions, analog chips responsible for transmission and energy supply, and Microcontroller Units (MCUs) which integrate computing, storage, and other functions onto a single chip. Their market shares constitute 22.41%, 22.41%, 10.79%, and 13.28% of the overall semiconductor market respectively, with logic and memory chips having relatively high proportions.

Non-integrated circuit semiconductor components (discrete devices, optoelectronic devices, sensors) account for 17% of the total semiconductor market share, which will be detailed in subsequent articles in this series.

Logic chips are integrated circuits that involve logical relationships and operate on binary principles to achieve computational and logical judgment functions. Common logic chips include CPUs (Central Processing Units), GPUs (Graphics Processing Units), ASICs (Application-Specific Integrated Circuits), and FPGAs (Field-Programmable Gate Arrays).

CPU: The Brain of Computers

CPUs are perhaps the most familiar logic semiconductor devices to the public, comprising arithmetic, control, and storage units, serving as the core of computer operations and control. First appearing in 1971, they are the most widely used main control chips in computer systems, applied in personal PCs and data terminal servers.

The global CPU industry is dominated by Intel and AMD, with domestic chip market shares around 1% in the personal PC sector and about 2% in the server sector. Leading companies in China's CPU industry include Loongson Technology, Hygon, and Zhaoxin.

GPU: A Boon for Gaming Enthusiasts

As data processing demands in personal PCs surged, CPUs found it increasingly challenging to keep up, leading to the emergence of GPUs as auxiliary processors.

GPUs are specialized processors designed for executing graphics computations in personal computers, workstations, gaming consoles, tablets, and smartphones.

Compared to CPUs, GPUs sacrifice some control units in favor of more computation units (arithmetic units), allowing them to execute large-scale homogeneous data computations, such as graphics rendering, at high densities. Users interested in gaming know that computers with powerful graphics cards run 3D game graphics more smoothly, with the GPU being the processor within the graphics card. GPUs excel in massive parallel computing, which is also essential for tasks like cryptographic cracking. Thus, besides image processing, GPUs are becoming increasingly involved in general computing.

Although CPUs have more storage and control units, making them more suitable for complex calculations, GPUs serve as significant complements to CPUs.

In design, there are notable differences between CPUs and GPUs due to their distinct missions. As described in a metaphor from NVIDIA architect Stephen Jones: "A GPU is a throughput machine, designed to work far beyond its runtime, like a train when not fully loaded; while a CPU is a latency machine, where thread-switching costs are substantial, aiming for a thread to run as quickly as possible."

Currently, the global GPU market is monopolized by NVIDIA, Intel, and AMD, capturing nearly 100% market share.

Domestically, Jingjia Micro holds promise for breakthroughs in the GPU sector. The company's JM9 series GPU successfully taped out in September 2020 and completed performance testing in November, aiding its expansion into civilian markets.

ASIC: Integrated Circuits for Specific Purposes

With the advent of smaller smart devices like smartphones and wearable tech, chip integration levels have increased, giving rise to Application-Specific Integrated Circuits (ASICs), characterized by user-involved design. ASICs often integrate multiple small-scale circuit modules, such as CPUs, GPUs, flash memory as well as Bluetooth, WIFI, etc., onto a single chip to meet system-level design needs. This approach is also known as System-on-Chip (SoC). It optimizes the entire circuit, reduces component count, shortens wiring, decreases size and weight, and enhances system reliability.

Smartphones are major application scenarios for ASICs. The well-known Kirin chips belong to this category. Huawei established an ASIC design center in 1991 for related research, which became the predecessor of HiSilicon, founded in 2004.

After enduring a year of sanctions, Huawei's HiSilicon announced on January 1: "2022, Moving Forward with 'Chip'." Cracks are where the light gets in, with Huawei Kirin's primary goal for 2022 being to overcome restrictions and restore normal Kirin chip supplies.

FPGA: Flexible Architecturally Programmable Chips

While ASICs optimize circuitry and reduce device sizes, they also increase design complexity and risk of tape-out failures.

This context led to the emergence of Field-Programmable Gate Arrays (FPGAs).

The standout feature of FPGAs is their namesake programmability in the field. Unlike CPUs, GPUs, or ASICs where functions are fixed post-manufacturing, FPGAs remain unfixed upon creation. Users can program FPGAs using software provided by FPGA vendors to configure functions, converting available modules into ones with specific desired functionalities.

FPGAs offer substantial flexibility, enabling users to alter chip logic functions in just minutes, easily navigating different business needs. Compared to ASIC solutions, FPGA solutions can reduce the risks associated with tape-out failures. Currently, China's FPGA market is dominated by Xilinx and Intel, commanding approximately 90% market share.

Anlu Technology leads domestically within the FPGA field, having been listed on the Science and Technology Innovation Board on November 12, 2021, with a current market share nearing 1%.

The Differences Between Logic Chips and Memory Chips

Logic chips and memory chips serve distinct purposes and exhibit fundamental differences, particularly in their transistor structures and operational modes, despite both being integral components of integrated circuits.

1.Fundamental Difference: The primary distinction between these two types of chips lies in the structure and working mode of their transistors. While they share some fundamental principles, their applications and functionalities differ significantly.

2.Size: When it comes to size, logic chips like Intel's CPUs currently operate around 20nm technology nodes, whereas memory chips, such as NAND Flash, have pushed closer to 10nm. This is measured by logic gate lengths; for example, the logic gate length (Lg) for 14/16nm FINFETs is smaller than the half-pitch of non-contact polysilicon in 2D NAND Flash. According to ITRS 2015, the former measures at about 15 nanometers, while the latter stands at 24 nanometers.

3.Nomenclature: In the semiconductor realm, naming conventions for process nodes (or technology nodes) vary between non-volatile memory (Flash) and logic products (MPU/ASIC). Non-volatile memory names its nodes based on the half-pitch of contact metal lines, whereas logic products use the half-pitch of non-contact polysilicon—despite the actual dimensions of source line (SL) and bit line (BL) being larger than the node number, the physical Lg in logic diagram is actually smaller than the node number.

4.New Structures: With the advent of new technologies like FINFET and 3D NAND, definitions around these points have evolved. For instance, in the 14/16nm FINFET technology mentioned earlier, the half-pitch of contact metal wires is actually 28nm, not the nominal 14/16nm. Meanwhile, the minimum array half-pitch for 3D NAND is approximately 80nm, which now defines its node.

5.Estimation: Discrepancies between nominal node numbers and real process parameters, alongside differing corporate nomenclatures, often lead to confusion. ASML provides an estimation formula that enables calculation based on each company's actual processes, with semiconductor industry typically using parameters close to nominal node numbers for calculations.

6.Technological Advancement: Currently, the structural differences between the two chip types, along with varied evaluation methodologies, make direct comparisons of technological advancement challenging. Each type progresses along its own path toward optimizing performance, and rather than declaring one more advanced than the other, it's apparent that they are tailored to meet different demands in their respective domains.