Level 1 - Absolute Beginner
Samsung showed a new memory chip at a big tech event called Hot Chips 2026. The event happened in late August 2026.
The new chip is called LPDDR5X-PIM. PIM means processing-in-memory. This means the chip can do math work inside itself.
Normal chips send data back and forth to a processor. This new chip does some of the work right where the data is stored. This makes it faster.
Samsung says the chip is up to three times faster for AI tasks. It can go into phones, computers, and small AI devices.
- chip
- a small piece of electronics that stores or processes information
- memory
- the part of a device that stores information
- processor
- the part of a computer that does the thinking and math
- data
- information used or stored by a computer
- AI
- artificial intelligence; computer programs that can learn and answer questions
- device
- a machine, like a phone or computer
- task
- a piece of work that needs to be done
- conference
- a big meeting where people share new ideas
Level 2 - Elementary
At the Hot Chips 2026 conference in late August, Samsung Electronics introduced a new kind of memory chip called LPDDR5X-PIM, which stands for processing-in-memory.
Instead of constantly sending data back and forth between the memory chip and a separate processor, this design lets certain calculations happen directly inside the memory chip itself. Samsung calls it the first LPDDR-based PIM design to become a real product in 2026.
The chip fits 16 gigabytes of memory across four dies inside a standard package, and its data transfer speed reaches 614 gigabytes per second. Samsung says it can be up to 3.01 times faster than regular LPDDR5X memory when running AI inference, which is when a trained AI model answers a new question.
A major advantage is that it can replace existing memory chips without requiring a completely new system design. Samsung is aiming this chip at smartphones, AI-enabled PCs, and edge AI devices, small devices that run AI locally instead of relying on a distant data center.
- calculation
- the process of using math to figure something out
- design
- the way something is planned or built
- gigabyte
- a unit used to measure a large amount of digital storage
- die
- a small piece of silicon that a chip's circuits are built on
- bandwidth
- how much data can move through a system in a certain amount of time
- inference
- when a trained AI model uses what it learned to answer a new question
- replace
- to put a new thing in the place of an old one
- local
- happening in a nearby place, rather than far away
Level 3 - Intermediate
At the Hot Chips 2026 conference, held around August 24 to 26, Samsung Electronics unveiled LPDDR5X-PIM, a memory chip built around the concept of processing-in-memory, which allows certain calculations to occur directly within the memory itself rather than requiring constant data movement between separate memory and processor components.
Samsung describes it as the world's first LPDDR-based PIM design being turned into an actual product in 2026. The chip embeds 16 processing-in-memory blocks inside its DRAM banks, alongside circuits dedicated to the repetitive arithmetic AI models rely on, plus a unit capable of handling both integer and floating-point calculations.
Built in a standard 561-ball package, the chip packs 16 gigabytes of memory across four dies within a single module, positioning it for use in servers, mobile phones, and PCs alike, and its bandwidth reaches 614 gigabytes per second. According to reports from Tom's Hardware and other outlets covering the conference, Samsung claims the chip is up to 3.01 times faster than standard LPDDR5X memory on AI inference workloads, alongside roughly eight times the bandwidth.
A key selling point is that LPDDR5X-PIM can serve as a drop-in replacement for existing LPDDR5 memory chips, avoiding the need for a redesigned system architecture. Samsung is positioning it for smartphones, AI-enabled PCs, and edge AI devices, a segment where high-bandwidth memory, the type favored in large AI data centers, tends to be too costly, power-hungry, and complex to deploy.
- embed
- to fix something firmly and deeply within a surrounding structure
- DRAM
- dynamic random-access memory, a common type of memory chip used to temporarily store data
- arithmetic
- the branch of mathematics dealing with basic operations like addition and multiplication
- integer
- a whole number, without a fraction or decimal part
- floating-point
- a way of representing numbers that include decimal points in a computer
- module
- a self-contained unit that can be combined with others to form a larger system
- workload
- the amount and type of work a system is required to perform
- architecture
- the fundamental structure and design of a computer system
Level 4 - Advanced
At the Hot Chips 2026 conference, convened around August 24 to 26, Samsung Electronics unveiled LPDDR5X-PIM, a memory architecture predicated on processing-in-memory, an approach that relocates certain computations into the memory chip itself rather than perpetually shuttling data between memory and a discrete processor.
Samsung bills it as the world's first LPDDR-based PIM design being commercialized in 2026. The chip embeds 16 processing-in-memory blocks within its DRAM banks, complemented by circuitry tailored to the repetitive multiply-accumulate arithmetic that underpins AI models, alongside a unit capable of both integer and floating-point computation.
Housed in a standard 561-ball package, the chip integrates 16 gigabytes of memory across four dies within a single module, a configuration intended to span servers, mobile handsets, and PCs alike, and it achieves a bandwidth of 614 gigabytes per second. Per reporting from Tom's Hardware and other outlets covering Hot Chips 2026, Samsung asserts the chip delivers up to 3.01 times the throughput of standard LPDDR5X on AI inference workloads, the process by which an already-trained model responds to new input, while furnishing roughly eightfold the bandwidth.
Its principal commercial advantage lies in its drop-in compatibility: LPDDR5X-PIM can supplant existing LPDDR5 memory without necessitating a redesigned system architecture. Samsung is positioning it for smartphones, AI-enabled PCs, and edge AI devices, a segment in which high-bandwidth memory, the variant favored in large-scale AI data centers, remains prohibitively costly, power-intensive, and architecturally complex to deploy.
- predicated
- based on or founded upon a particular idea or principle
- discrete
- separate and distinct, not connected to other things
- commercialize
- to develop something into a product that can be sold in the market
- multiply-accumulate
- a common computing operation that multiplies two numbers and adds the result to a running total
- throughput
- the amount of work or data a system can process in a given period
- supplant
- to take the place of something, especially through displacement
- prohibitively
- to a degree that makes something impractical, especially due to cost
- architecturally
- relating to the fundamental structure or design of a system
Talk about this article with KatePractice this story's vocabulary in a real spoken conversation with your AI tutor