Skip to content
Apex
ApexProductsAether for semiconductors

Aether for semiconductors.

AI that retires the EDA stack.

Aether for semiconductors is what the incumbent EDA stack collapses into when the model is the substrate. Specialist agents drive a complete chip flow — 3DIC architecture, RTL through signoff, multiphysics, manufacturing, silicon lifecycle — and a comprehensive silicon-IP catalog comes with the platform.

Agents
17
IP blocks
107
Flow
RTL → GDS
Tape-outs
In production
Coverage

Twelve agents you'd hire. One flow.

EDA is a duopoly of mature stacks with mature lock-in. We don't displace every niche tool, and we don't try. But the spine of the flow — architecture, implementation, verification, signoff, manufacturing — lives behind a single set of contracts.

01

3DIC architecture

Chiplet partitioning, interposer planning, thermal-aware floorplanning for stacked dies. UCIe and BoW interconnect modelling, die-to-die signal integrity, redistribution-layer routing.

02

Physical implementation

Synthesis, placement, clock-tree synthesis, routing, congestion management. RTL handed in, GDS handed out. Aware of power, congestion and design-for-manufacturing constraints from the first iteration.

03

IP integration

Drives a comprehensive silicon-IP catalog. Pin-accurate, timing-aware integration; auto-generated wrappers, integration tests, and back-annotated SDF.

04

Verification orchestration

Functional verification, formal property checking, emulation orchestration, coverage closure, regression triage. UVM testbench generation from spec.

05

Multiphysics signoff

Coupled thermal, electromagnetic, mechanical analysis — signoff-grade. IR-drop, EM, ESD, thermal hot-spotting, package warpage.

06

Manufacturing optimisation

DFM rules, hot-spot detection, yield analysis, OPC, mask synthesis preparation — pinned to your foundry's PDK across the leading-edge nodes.

07

Silicon lifecycle

Post-tape-out monitoring, in-field telemetry, reliability and aging models. Closes the loop between deployed silicon and the next design.

08

Processor configuration

ARC-V (RISC-V) and NPU configuration. Per-workload PPA tuning, custom instruction extensions, vector/matrix unit sizing.

09

SoC interconnect

NoC topology, bus contention, QoS shaping, virtual-channel allocation — for the workload, not for a template.

10

Security

Side-channel hardening, fault-injection analysis, secure-boot, root-of-trust, key management, post-quantum primitives baked into the flow.

11

Reliability & aging

BTI, HCI, EM, TDDB, stress-migration — modelled across PVT corners and projected to end-of-life under your workload mix.

12

AI design-space optimisation

DSO.ai-class search across PPA targets — but driven by Aether, not by a black-box optimiser. Decisions are explainable and reproducible.

The flow

From PRD to silicon, under one agent.

Six phases of the chip flow, end-to-end. Every phase is observable, every hand-off is auditable, and every ECO round-trips through the model rather than dying in a folder of TCL.

  1. 01

    Architecture

    From PRD to a partitioned, budgeted SoC: block-level area, power and frequency budgets, NoC topology, IP selection, package and 3DIC choices.

  2. 02

    Implementation

    RTL coding, synthesis, place-and-route, clock-tree, timing closure. Hand-off to signoff is automated; ECOs round-trip through the model.

  3. 03

    Verification

    UVM testbench generation, formal proof of safety properties, gate-level simulation, gate-level coverage, regression triage with explainable failure traces.

  4. 04

    Signoff

    Static timing across PVT corners, power and IR analysis, EM, thermal, ESD, package warpage — all driven by the same model that did implementation.

  5. 05

    Tape-out

    DFM, hot-spot fix, OPC prep, mask synthesis hand-off to the foundry. The model knows what the foundry rules forbid before you ask.

  6. 06

    Lifecycle

    In-field telemetry from deployed silicon flows back into the next design. Reliability models update; the second tape-out is better than the first.

Stack it replaces

In place of three vendors.

These are the tools in a typical fabless or IDM EDA stack today. Aether for semiconductors replaces the spine and the seat-licensing tax that comes with it. Niches — custom analog cells, foundry-specific signoff — stay where they are.

Logic synthesis
Physical implementation
Static timing signoff
Functional verification
AI-driven PPA search
Formal equivalence checking
Block synthesis
Place-and-route
Multi-corner timing
Power + IR signoff
RTL simulation
ML-driven design exploration
Logic equivalence
Manufacturing test
DRC + LVS + DFM
Verification IP
Silicon-IP catalog

107 blocks, shipped with the platform.

The IP integration agent stitches these into your floorplan without the per-block licence dance. Each block has a verification suite, a power model, and integration tests that ship with it.

ARC-V RISC-V cores (32, 64-bit)
NPU configurations (INT8, FP16, FP8)
USB / PCIe / CXL controllers
DDR / LPDDR / HBM controllers
Ethernet MAC + SerDes
Display / camera serial
Crypto + secure boot
Power / clock infrastructure
Mixed-signal / data converters
Sensor interfaces (I3C, SPI, I2C)
Memory compilers
Standard-cell libraries
Process nodes

Pinned to real PDKs.

The model is calibrated against the rules and constraints of the foundry node you actually tape out on. Every supported PDK has a validated regression suite.

Leading-edge 3nm
Mainstream 5nm
Cost-optimised 7nm
Leading-edge 3nm gate-all-around
Sub-2nm angstrom node
Mature 12nm low-power
FD-SOI 22nm
Specialty automotive nodes
Vs the EDA spine

RTL to signoff, one database.

The EDA stack is a dozen point tools and as many file hand-offs. Aether drives the whole flow from one model and one database — DFM-aware from the first iteration.

✓RTL synthesis
✓Place & route
✓Static timing signoff
✓Functional / formal verification
✓Multiphysics signoff (power/IR/thermal)
✓DFM / OPC / manufacturing
✓3DIC / chiplet architecture
✓RTL → GDS, one database
Why the cycle shortens

Quarters of tape-out cycle, compressed to weeks.

The slowest part of taping out a chip isn't any one solve — it's the script handoffs, the corner sweeps and the ECO loops. Aether collapses each of those into the autonomous agent flow on one database.

Stage
Today
With Aether
Why
Spec to RTL
8–16 weeks
1–2 weeks
Micro-architectural design-space exploration as part of the autonomous study. The planner proposes RTL variants under your power, area and timing budget — you pick.
RTL to signoff-clean P&R
12–20 weeks
2–4 weeks
Synthesis, place-and-route, timing, power and EM/IR run on one database. No more weeks of script translation between flow stages.
Closing timing / power / EM
weeks of ECOs
hours of optimiser passes
Eval-gated optimisation loops drive each metric back into spec. The agent that found the violation is the agent that proposes the fix.
Tape-out hand-off
days
hours
GDS, abstracts, sign-off reports and the foundry-pack metadata bundle auto-assembled. Hash-verified against the signoff snapshot.
Performance at iso-area
1.06×

vs. a leading commercial EDA flow on the public 7nm predictive PDK. ≤0.4% WNS gap. Reproduction scripts in the research log.

Tape-out cycle time
−40%

Programme-level compression measured at the first fabless customers. The dominant gain is fewer ECOs and faster signoff closure.

EDA license servers
0

Fabless customers running Aether for semiconductors on the published flow report no incumbent-EDA license consumption for their primary signoff.

AI scientists, not chatbots

Aether is an autonomous chip designer.

Not a chat assistant for your EDA flow — a foundation model that takes a spec and returns a GDS. Plans the floorplan, runs the agents, clears signoff, hands off to the foundry.

  • 01

    Reads the spec

    Ingests the micro-architecture, the power and area budget, the foundry process and your IP catalog. Decomposes the brief into a floorplan-aware study tree.

  • 02

    Picks the right flow

    Knows when synthesis margin is enough and when physical-aware synthesis is required. Knows which library corner pairs with which signoff. The senior physical-design engineer's intuition encoded.

  • 03

    Runs synthesis, P&R, signoff

    Specialist production agents execute the flow as a DAG. Long routing runs check-point. Signoff agents share the same database the placement agent wrote to.

  • 04

    Validates against signoff

    STA, IR, EM, DRC, LVS — all gates the model has to clear before any artefact reaches you. Drift halts the run and surfaces the offending net.

  • 05

    Writes the signoff pack

    Auto-written signoff reports with WNS / TNS distributions, IR-drop maps, EM violations and the GDS hash. Ready for tape-out review.

  • 06

    Closes the foundry loop

    Foundry-pack bundles, abstract-cell hand-off, version-pinned process kits — assembled and signed by the model version. Foundries get a clean drop, not a folder of one-offs.

Where this leads

A foundation model that tapes out, not just suggests RTL — the substrate for general silicon intelligence.

Aether reasons in RTL, logic, timing and physics — the substrate of digital silicon. We are moving toward general intelligence through the disciplines that have to face physical reality. The model that signs off your chip today is the model that designs the next one tomorrow.

Validation

Numbers against commercial baselines.

We benchmark against the commercial EDA stacks themselves, on published cases. Reproduction scripts are in the research log.

Public open-cell-library timing
≤0.4%
WNS gap vs commercial baseline
Public 7nm predictive PDK
1.06×
perf at iso-area vs commercial flow
AI-driven PPA design-space search
Parity
convergence on shared bench
EM/IR signoff agreement
≤2.0%
current-density delta
Related research

The work behind this product.

We publish what we learn. The posts below are the substantive notes behind this product — methods, evaluations, case studies.

Bring it to your next tape-out.

Send us the spec and the schedule. We'll show you what Aether does with one block of your real design — under your foundry's PDK.