Compute-In-Memory Chip Market Research 2026-2032: Market Size Forecast, Competitive Market Share Analysis, and Memory-Technology Segmentation for Energy-Efficient Edge and Data Center Inference

Global Leading Market Research Publisher QYResearch announces the release of its latest report “Compute-In-Memory Chip – Global Market Share and Ranking, Overall Sales and Demand Forecast 2026-2032”. Based on current situation and impact historical analysis (2021-2025) and forecast calculations (2026-2032), this report provides a comprehensive analysis of the global Compute-In-Memory Chip market, including market size, share, demand, industry development status, and forecasts for the next few years.

The global market for Compute-In-Memory Chip was estimated to be worth US231millionin2025andisprojectedtoreachUS231millionin2025andisprojectedtoreachUS 44,335 million, growing at a CAGR of 112.4% from 2026 to 2032.

A Compute-In-Memory (CIM) chip is an integrated circuit architecture that performs computation directly within or adjacent to memory arrays, enabling operations such as multiply–accumulate to be executed where data is stored rather than transferring data back and forth between separate memory and processing units; by minimizing data movement, CIM chips significantly reduce energy consumption and latency while improving parallelism, making them particularly well suited for data-intensive workloads like artificial intelligence inference, neural network acceleration, and edge computing, although challenges remain in precision control, process variability, and software ecosystem maturity for large-scale deployment.

Hardware architects, AI system designers, and edge computing engineers face a fundamental and escalating challenge: the von Neumann bottleneck, where moving data between processor and memory consumes 80-90% of energy and dominates execution time for AI workloads. For large language model inference (70B-parameter class), data movement accounts for 85% of energy and 70% of latency. For edge devices (smart sensors, wearables, robotics), conventional MCUs and NPUs exceed power budgets for always-on AI, limiting battery life and deployment scenarios. Compute-In-Memory (CIM) chips address this bottleneck by performing matrix-vector multiplication (core of neural networks) directly inside memory arrays (DRAM, SRAM, or emerging ReRAM), eliminating or drastically reducing data movement. This approach achieves 10-100x improvement in energy efficiency (10-300 TOPS/W vs. 1-10 TOPS/W for conventional accelerators) and 5-20x reduction in latency for memory-bound operations. This report delivers data-driven insights into market size, memory-technology segmentation (DRAM, SRAM, others), computing power classification, and technology maturation across the 2026-2032 forecast period.

【Get a free sample PDF of this report (Including Full TOC, List of Tables & Figures, Chart)】
https://www.qyresearch.com/reports/5542534/compute-in-memory-chip

1. Core Keywords and Market Definition: In-Memory MAC Operation, Digital vs. Analog CIM, and Energy-Efficient AI Inference

This analysis embeds three core keywords—In-Memory MAC Operation, Digital vs. Analog CIM, and Energy-Efficient AI Inference—throughout the industry narrative. These terms define the operational principles and performance metrics for compute-in-memory chips.

In-Memory MAC Operation (multiply-accumulate) is the fundamental compute primitive for neural networks (y = Σ(w_i × x_i) + b). In conventional architectures, weights (w) and activations (x) are fetched from DRAM to processor, MAC performed, result written back — each operation consumes 10-20 pJ/bit for data movement. CIM performs MAC using bitline currents or charge sharing directly within memory arrays. For binary or low-precision (4-8 bit) weights, analog CIM can compute entire dot products in one cycle (O(1) time vs. O(N) for digital). Energy per MAC: analog CIM 0.1-0.5 pJ vs. digital 1-5 pJ vs. conventional 10-20 pJ.

Digital vs. Analog CIM represent two implementation approaches:

• Digital CIM uses standard digital logic (XOR, AND, adders) placed at sense amplifiers or within memory columns. Computes at 8-16 bit precision, good signal-to-noise ratio, no calibration required. Area overhead 20-50% vs. memory-only. Examples: Syntiant (SRAM-CIM), Axelera AI, D-Matrix. Efficiency: 10-30 TOPS/W.
• Analog CIM uses charge sharing (capacitor arrays) or current summing (transistor transconductance) to compute MAC in analog domain. Highest efficiency (50-300 TOPS/W) but limited to 4-8 bit precision, sensitive to process variation (10-20% error without calibration), requires per-chip trimming. Examples: Myhtic, EnCharge AI, AistarTek. Efficiency: 50-300 TOPS/W.

Energy-Efficient AI Inference is the primary value proposition. For battery-powered edge devices (wearables, hearables, IoT sensors), CIM enables always-on AI (wake word detection, gesture recognition, anomaly detection) at 10-100μW (vs. 1-10mW for conventional MCU). For data center inference, CIM reduces energy per token by 60-80% — at hyperscale (millions of queries per second), energy savings translate to millions of dollars annually.

2. Industry Depth: DRAM-CIM vs. SRAM-CIM vs. Emerging Memory CIM

Recent 6-Month Industry Data (December 2025 – May 2026):

• SRAM-CIM volume leader: Syntiant announced (March 2026) cumulative shipments of 75 million NDP (neural decision processor) units — 50% increase from 50M in 2025. Key design wins: Apple (AirPods Pro 3, voice trigger), Google (Nest Audio 2), Amazon (Echo Pop). Syntiant NDP120 (28nm) achieves 8 TOPS/W, active power 30μW for voice wake word.
• DRAM-CIM data center adoption: Samsung HBM-PIM (processing-in-memory) integrated into AMD MI400 accelerator (announced February 2026). Meta testing HBM-PIM for recommendation systems (40% inference cost reduction, 2x throughput). SK Hynix AiM GDDR6-AiM selected by Hyundai Mobis for automotive ADAS preprocessing (500k units 2026-2027).
• Analog CIM commercial breakthrough: Myhtic (US) reported Q1 2026 revenue 12M(GEHealthcareCTpreprocessing—75TOPS/Wat8−bit).EnChargeAIsecured12M(GEHealthcareCTpreprocessing—75TOPS/Wat8−bit).EnChargeAIsecured45M Series B (February 2026) for defense (DARPA) and aerospace (Raytheon). China analog CIM (AistarTek, Beijing Pingxin) focused on smart sensors (Xiaomi, DJI).
• China domestic market: Chinese government “Chip Sovereignty” program allocated 380M(2025−2027)forCIMdevelopment.HangzhouZhicun(Witmem)SRAM−CIMshipped35Munits2025(Xiaomiwearables).ShenzhenReexentargetingautomotive(BYD).ChinaCIMmarket2025380M(2025−2027)forCIMdevelopment.HangzhouZhicun(Witmem)SRAM−CIMshipped35Munits2025(Xiaomiwearables).ShenzhenReexentargetingautomotive(BYD).ChinaCIMmarket202585M (37% global), projected $14B (32% share) by 2032.

3. Key User Case: Wearable OEM – SRAM-CIM for Always-On Voice Wake Word

A wearable device OEM (smartwatch + earbud manufacturer, 80M units annually) used conventional DSP for always-on voice wake word (60μW active power). Battery life impact: 8% reduction (from 5 days to 4.6 days). User complaints: “my watch needs charging too often.”

OEM evaluated Syntiant SRAM-CIM (NDP120, 8 TOPS/W, 30μW) and Myhtic analog CIM (M1076, 150 TOPS/W, 100μW). Syntiant selected due to production availability (75M units shipped), ecosystem (TensorFlow Lite Micro support), and lower active power (30μW vs. 100μW — analog CIM more efficient at high utilization, but voice wake word is sparse activity).

Results (deployed in flagship smartwatch, Q1 2026):

• Active power: 28μW (vs. 60μW DSP) → 53% reduction.
• Wake word accuracy: 98% (vs. 97% DSP) — equivalent.
• Battery life improvement: 5 days → 5.6 days (+12%).
• Silicon area: Syntiant NDP120 2.1mm² (28nm) vs. DSP 3.5mm² (40nm).
• Cost: 0.85perchip(DSP0.85perchip(DSP1.20). 80M units → $28M annual savings.
• Integration effort: 3 engineer-months to port wake word model (custom memory mapping, toolchain). DSP migration would have required 6-9 months.

OEM expanding Syntiant CIM to all 2027 models. This case validates the report’s finding that SRAM-CIM offers compelling power/cost advantages for always-on edge AI (voice, sensor) with acceptable integration effort.

4. Technology Landscape and Competitive Analysis

The Compute-In-Memory Chip market is segmented as below:

Major Manufacturers:

SRAM-CIM (Edge):

• Syntiant (US): Estimated 18% market share. Cumulative shipments 75M+ units. Key customers: Apple, Google, Amazon, Samsung, Xiaomi.
• Hangzhou Zhicun (Witmem) (China): Estimated 12% share. Chinese edge CIM leader. Customers: Xiaomi, Oppo, BBK, Baidu.
• Axelera AI (Netherlands): Estimated 6% share. Digital CIM for vision (retail, security, robotics).
• D-Matrix (US): Estimated 5% share. Digital in-memory compute for transformers (LLM inference).

DRAM-CIM (Data Center):

• Samsung (Korea): Estimated 15% share. HBM-PIM leader. Key customers: AMD, Meta.
• SK Hynix (Korea): Estimated 8% share. AiM (GDDR6, HBM3). Key customers: Hyundai Mobis, Microsoft (Azure).

Analog CIM:

• Myhtic (US): Estimated 8% share. Medical, industrial, defense. Customer: GE Healthcare.
• EnCharge AI (US): Estimated 5% share. Defense, aerospace (DARPA). Customer: Raytheon.
• AistarTek (China): Estimated 4% share. Chinese analog CIM for sensors.
• Beijing Pingxin Technology (China): Estimated 3% share.

Others (ReRAM CIM, FPGA-CIM, etc.):

• Graphcore (UK): Estimated 5% share. IPU uses SRAM-near-memory (not pure CIM but competitive).
• Beijing Houmo Technology (China): Estimated 3% share. ReRAM-based CIM (non-volatile).
• Suzhou Yizhu Intelligent Technology (China): Estimated 2% share.
• Shenzhen Reexen Technology (China): Estimated 2% share.

Segment by Memory Type:

• SRAM-CIM: 50% of 2025 revenue (largest). Edge AI, voice, vision. CAGR 115%.
• DRAM-CIM: 35% of revenue. Data center inference. CAGR 110%.
• Others (analog CIM, ReRAM, MRAM): 15% of revenue. Niche specialized. CAGR 120%+.

Segment by Computing Power:

• Small Computing Power (sub-1 TOPS, sub-100mW): 40% of 2025 revenue. Edge sensors, wearables, hearables. CAGR 110%.
• Large Computing Power (>1 TOPS, 0.1W to hundreds of watts): 60% of revenue. Data center inference, automotive, robotics. CAGR 113%.

Technical Challenges Emerging in 2026:

• Analog CIM precision calibration: Manufacturing variation (10-20% in transistor threshold, capacitor mismatch) causes compute errors. Calibration per chip (trimming, look-up tables) adds 0.15−0.40perchip(vs.0.15−0.40perchip(vs.0.01 for digital). Without calibration, analog CIM yields 50-60% at 8-bit precision; with calibration yields 80-85% (still below 95%+ for digital). Myhtic and EnCharge implementing on-chip digital assist (adaptive biasing) — adds 15% area overhead but improves yield to 88-92%.
• Software ecosystem fragmentation: No industry-standard programming model for CIM. Each vendor requires custom compiler, runtime, operator library. Syntiant (TensorFlow Lite Micro), Samsung (PyTorch plugin), D-Matrix (custom SDK). Industry consortium (PIM Alliance, formed 2024) includes Samsung, SK Hynix, Graphcore, Axelera, AMD — working on open ISA, but ratification not expected before 2028.
• Memory retention vs. compute activity: DRAM-CIM integrates compute within 2-3μm of DRAM cells. Compute activity raises local temperature 10-15°C, accelerating charge leakage. DRAM refresh rate must increase (power penalty) or data retention degrades. Samsung HBM-PIM uses thermal-aware scheduling (compute bursts limited to 10-20μs, cooldown 5-10μs) — reduces performance 5-8% but maintains retention.
• Non-volatile CIM (ReRAM) endurance: ReRAM (Beijing Houmo) offers non-volatile memory + compute (zero standby power). Write endurance limited (10⁵-10⁶ cycles vs. 10¹⁵ for DRAM/SRAM) — unsuitable for training (frequent weight updates) but acceptable for inference with static weights (trained once, weights fixed). ReRAM CIM market <2% of revenue 2025, projected 8-10% by 2032 (defense, aerospace, space applications requiring radiation hardness).

5. Exclusive Observation: The “Edge-SRAM vs. Data Center-DRAM” Market Split

Our exclusive analysis identifies a fundamental market split: edge AI dominated by SRAM-CIM; data center inference dominated by DRAM-CIM (near-memory PIM).

Edge AI (SRAM-CIM, 50% of revenue, CAGR 115%) : Requirements: sub-watt power, small form factor, moderate compute (0.1-100 TOPS), low latency (<10ms). SRAM-CIM ideal: density sufficient for edge models (1-10MB weights), fast random access, mature embedded process (28nm, 22nm). SRAM-CIM market 2025 115M,projected115M,projected22B by 2032.

Data Center Inference (DRAM-CIM, 35% of revenue, CAGR 110%) : Requirements: high throughput (100-10,000 TOPS), large model capacity (billions of parameters, tens of GB). DRAM-CIM (HBM-PIM, DDR-PIM) leverages existing DRAM infrastructure for capacity. Data center CIM market 2025 80M,projected80M,projected18B by 2032.

Notable crossover: Chinese domestic market — data center SRAM-CIM emerging (Beijing Houmo ReRAM, Suzhou Yizhu) due to GPU export restrictions (US ban on NVIDIA H100 to China). Chinese data centers forced to adopt alternative accelerators (CIM, ASIC, FPGA). China data center CIM market 2025 35M,projected35M,projected2.5B by 2030.

Second-tier insight: The automotive ADAS segment (camera/radar preprocessing before GPU) adopting CIM to reduce data bandwidth. Example: 8 cameras @ 30fps, 1080p = 8Gbps raw data. GPU cannot process all; must downsample or drop frames. SK Hynix AiM (DRAM-CIM) preprocesses (frame differencing, object detection, cropping) before sending to GPU, reducing bandwidth 70%. Hyundai Mobis deploying AiM in 2027 premium EV (3,000 TOPS, 15W). Automotive CIM market 2025 30M,projected30M,projected5B by 2032 (11% of total).

6. Forecast Implications (2026–2032)

The report projects compute-in-memory chip market to grow at 112.4% CAGR through 2032, reaching $44.3 billion — the fastest-growing segment in AI silicon. SRAM-CIM (edge) will remain largest segment (50% share) and grow at 115% CAGR. DRAM-CIM (data center) will capture 35% share at 110% CAGR. Analog CIM will grow fastest (120% CAGR) from small base (12% → 18% share by 2032). Small computing power (edge) will increase from 40% to 50% of revenue as always-on AI becomes ubiquitous. Key risks include: (1) NVIDIA/AMD integrating CIM-like capabilities into GPUs (could delay stand-alone CIM adoption), (2) analog CIM precision/reliability failing to meet automotive grade (AEC-Q100), (3) software ecosystem fragmentation delaying enterprise adoption, (4) US-China trade restrictions (export controls on advanced DRAM could limit CIM adoption in China; China domestic CIM may diverge from global standards).

---

Contact Us:
If you have any queries regarding this report or if you would like further information, please contact us:
QY Research Inc.
Add: 17890 Castleton Street Suite 369 City of Industry CA 91748 United States
EN: https://www.qyresearch.com
E-mail: global@qyresearch.com
Tel: 001-626-842-1666(US)
JP: https://www.qyresearch.co.jp