Pascari aiDAPTIV™ Cache Memory is the flash-based capacity tier in the aiDAPTIV memory architecture.
It works with aiDAPTIV Middleware to retain eligible AI state when GPU memory, unified memory, or system memory is constrained. Depending on the workload and runtime, that state can include KV cache, less-active MoE experts, model data, and training state.
This is not ordinary overflow storage. aiDAPTIV Cache Memory remains part of the active AI workflow, retaining useful state in a larger tier instead of discarding it and rebuilding it later.
aiDAPTIV Cache Memory is not only about flash capacity.
The SSD controller and firmware influence how the drive handles sustained writes, latency variation, wear management, data placement, recovery behavior, and contention between workloads. Those characteristics matter when the SSD is used as an AI memory tier rather than passive storage.
Phison develops the controller, firmware, NAND configuration, and aiDAPTIV middleware stack as part of the broader solution. That gives the platform a path to align SSD behavior with the sustained data movement and endurance demands of AI cache workloads.
The goal is not to make flash behave like GPU memory. The goal is to provide a durable, predictable lower tier that can retain useful AI state when fast memory is constrained.
AI memory traffic is different from ordinary file storage.
A general-purpose SSD is usually sized and tuned for operating systems, applications, user files, and intermittent storage activity.
An AI cache tier may repeatedly retain, update, evict, restore, and overwrite data as the workload moves through the memory hierarchy.
That can include long-context KV cache, retrieved RAG material, inactive MoE experts, model-layer data, activations, and other training state.
Endurance is therefore part of the architecture, not a secondary specification.
aiDAPTIV Cache Memory is designed for sustained AI cache activity using high-endurance flash, specialized SSD design, and advanced NAND correction. Product-level endurance and warranty specifications vary by SKU.
aiDAPTIV Cache Memory is available in two deployment models:
Dedicated aiDAPTIV Cache Memory SSDs
A dedicated aiDAPTIV Cache Memory SSD gives the AI workload its own drive capacity.
This configuration is intended for systems with an available M.2, U.2, E1.S, or E3.S slot, where AI cache activity should remain separate from the operating system, application files, and user data.
Dedicated aiDAPTIV Cache Memory is suited to larger retained working sets, including long-context inference, repeated RAG or coding context, larger MoE expert pools, and memory-constrained fine-tuning.
Product family | Interface | Form factors | Capacity shown in roadmap |
|---|---|---|---|
AI100E | PCIe 4.0 | M.2 2280, U.2, E1.S | 320 GB to 2 TB |
AI200E | PCIe 5.0 | U.2, E3.S | 2 TB to 8 TB |
AI250E | PCIe 5.0 | M.2 2280 | 2 TB |
AI20E | PCIe 5.0 | M.2 2280 | 320 GB and 1 TB |
Product family
AI100E
Interface
PCIe 4.0
Form factors
M.2 2280, U.2, E1.S
Capacity shown in roadmap
320 GB to 2 TB
Product family
AI200E
Interface
PCIe 5.0
Form factors
U.2, E3.S
Capacity shown in roadmap
2 TB to 8 TB
Product family
AI250E
Interface
PCIe 5.0
Form factors
M.2 2280
Capacity shown in roadmap
2 TB
Product family
AI20E
Interface
PCIe 5.0
Form factors
M.2 2280
Capacity shown in roadmap
320 GB and 1 TB
A hybrid configuration combines conventional storage and aiDAPTIV Cache Memory in one drive.
The operating system, applications, and user files use the storage portion. aiDAPTIV uses the assigned cache memory capacity for eligible AI state.
This is a practical fit for AI PCs, laptops, mini PCs, compact desktops, and other systems where a second drive slot may not be available.
Product family | Interface | Form factors | Capacity shown in roadmap |
|---|---|---|---|
AI20EH | PCIe 5.0 | M.2 2280 | 768 GB storage + 85 GB cache 1.6 TB storage + 128 GB cache 3.3 TB storage + 256 GB cache |
Product family
AI20EH
Interface
PCIe 5.0
Form factors
M.2 2280
Capacity shown in roadmap
768 GB storage + 85 GB cache
1.6 TB storage + 128 GB cache
3.3 TB storage + 256 GB cache
Configuration | Use when: | Typical systems |
|---|---|---|
Dedicated aiDAPTIV Cache Memory SSD | System has an available drive slot and AI workload needs large retained-cache capacity without competing with operating-system or user-storage activity | Discrete-GPU desktops, workstations, edge systems, and servers |
Hybrid Storage + Cache SSD | Drive slots are limited and the system needs both conventional storage and aiDAPTIV Cache Memory in the same M.2 footprint | Laptops, mini PCs, compact desktops, and integrated or unified-memory systems with limited drive capacity |
Configuration
Dedicated aiDAPTIV Cache Memory SSD
Use when:
System has an available drive slot and AI workload needs large retained-cache capacity without competing with operating-system or user-storage activity
Typical systems
Discrete-GPU desktops, workstations, edge systems, and servers
Configuration
Hybrid Storage + Cache SSD
Use when:
Drive slots are limited and the system needs both conventional storage and aiDAPTIV Cache Memory in the same M.2 footprint
Typical systems
Laptops, mini PCs, compact desktops, and integrated or unified-memory systems with limited drive capacity
The right choice depends on the target platform, available drive bays or slots, runtime support, model and context requirements, and the amount of cache memory the workload needs.
English English 
繁體中文
简体中文
日本語
