2. Information Classification: General
December 8-10 | Virtual Event
Building an Open Control Stack for Quantum Computers
using RISC-V Ecosystem
Anastasiia Butko
Research Scientist
Lawrence Berkeley National Laboratory
#RISCVSUMMIT
6. Information Classification: General
QUANTUM SYSTEM
Quantum computer ≠ Quantum Processing Unit
Quantum computer = Quantum Processing Unit + Control Hardware
User
Quantum Processing Unit
Cryogenic Control Architecture for Large-Scale Quantum Computing, J. M. Hornibrook et al., Phys. Rev. Applied 3, 024010,
(2015)
8. Information Classification: General
• have been an active area of study long before any physical implementation of
quantum computers became available
• recent research and development bring the awareness of hardware to the high levels
(NISQ applications), but abstraction and modularity is crucial for scalability
ABSTRACTION LEVELS
Awareness across the Stack
9. Information Classification: General
• have been an active area of study long before any physical implementation of
quantum computers became available
• recent research and development bring the awareness of hardware to the high levels
(NISQ applications), but abstraction and modularity is crucial for scalability
ABSTRACTION LEVELS
Awareness across the Stack
• operate at the level of physical pulses with very limited or no notion about software
12. Information Classification: General
CONTROL UNIT
Software vs. Hardware Implementation
Software Control
• Everything in one place
• Easier to implement and modify
• NISQ experiments
• Long latency
• Limited bandwidth
• Limited flexibility (bit granularity)
• Poor scalability
S e q u e n c e
B a n d w i d t h
L a t e n c y 1 L a t e n c y 2
M E M
✅
✅
✅
❌
❌
❌
❌
L O A D
13. Information Classification: General
Q A S M
L a t e n c y 1 L a t e n c y 2
M E M
Hardware Control
• Reduced latency
• Fast feedback
• Extended functionality in-place
• Potential scalability
• Separation of concepts
• Beyond-NISQ
• Requires specialized solution
• Harder to implement and modify
• Requires commercial support
❌
❌
✅
✅
✅
✅
✅
✅
CONTROL UNIT
Software vs. Hardware Implementation
❌
14. Information Classification: General
• Quantum Instruction Set Architecture (QUASAR)
• extension to RISC-V ISA (open-source, modular, active community, eco-system)
• supports quantum operations, timing control, etc.
• transparent, adaptable, open
• New tools in hardware design and evaluation
• high-level HDL (Chisel)
• modular and parametrizable modules
Requires specialized solution
Harder to implement and modify
HARDWARE APPROACH
Building an efficient software-hardware interface
✅
✅
16. Information Classification: General
QUASAR Extensions
Different approaches
• Greenfield extension approach
• Tightly integrated with the processor micro-architecture
• High-level of customization
• May cause conflicts with other extensions
• Requires compiler extension
• RoCC co-processor adaptation
• Existing software support
• Modular approach easily interactable into a big system
• Lower-level of customization and flexibility
P r o c e s s o r
C o r e
L 1 I L 1 D
R I S C - V R o c k e t R o C C
m r q
m r s
c m d
r e s p
b u s y
Q U A S A R
D E C
W B
M E M
Q I
T S R
M S R
Q
P
U
17. Information Classification: General
QUASAR Extensions
Different approaches
• Greenfield extension approach
• Tightly integrated with the processor micro-architecture
• High-level of customization
• May cause conflicts with other extensions
• Requires compiler extension
• RoCC co-processor adaptation
• Existing software support
• Modular approach easily interactable into a big system
• Lower-level of customization and flexibility
P r o c e s s o r
C o r e
L 1 I L 1 D
R I S C - V R o c k e t R o C C
m r q
m r s
c m d
r e s p
b u s y
Q U A S A R
D E C
W B
M E M
Q I
T S R
M S R
Q
P
U
• Quantum ISA Spec evaluation
• Metrics and methodology
• Comparison against existing solutions
• Scalability and performance projections
Understanding Quantum Control Processor Capabilities and Limitations through Circuit
Characterization, A. BUTKO et al., IEEE International Conference on Rebooting Computing
(ICRC), December 2020
https://arxiv.org/abs/1909.11719
18. Information Classification: General
System Architecture Implementation
From core to system architecture
• Using Freedom Platform (SiFive) for system integration on VC707 FPGA
19. Information Classification: General
• Using Freedom Platform (SiFive) for system integration on VC707 FPGA
• Run Linux
• Remote communication
• Periphery support
• Security
System Architecture Implementation
From core to system architecture
20. Information Classification: General
• Using Freedom Platform (SiFive) for system integration on VC707 FPGA
• RoCC Interface
• Pulse Library Table (PLT)
• Additional functionalities
(loading Waveform Memory
for calibration, readout post-
processing )
System Architecture Implementation
From core to system architecture
21. Information Classification: General
• Using Freedom Platform (SiFive) for system integration on VC707 FPGA
• Analog backend
• Phase manipulation
• Fast readout
• Timing control
System Architecture Implementation
From core to system architecture
22. Information Classification: General
High-level circuit
QUASAR library
Linux OS
int main() {
qReset(); //reset hardware resources to ZERO
x90i(1); //apply the X90 immediate gate on qubit 1
tsi(5); //advances the time stamp on 5 cycles
y90i(0); //apply the Y90 immediate gate on qubit 0
return 0;
}
#define XCUSTOM_ACC 0
#define q_DO_X90 1
#define x90i(qid)
ROCC_INSTRUCTION_0_R_R(XCUSTOM_ACC, qid, 0, q_DO_X90);
Software Stack
From high-level circuit description to Linux executable
Control Pulses generated by
the Analog Backend
23. Information Classification: General
FUTURE WORK
• How scalable are our approaches?
• QUASAR supports up to 512 qubits
• Quantum vector extension (qV) to express single-instruction multiple-qubits multiple-gates parallelism
• Multiple QUASAR co-processors, multiple FPGAs
• Bit manipulation for error-correction code
• Cool technologies
• Cryo-CMOS: energy dissipation, noise
• Superconducting: how the micro-architecture will change?
… and much more
24. Information Classification: General
Advanced Quantum Testbed @ LBNL
Superconducting Quantum Processors at the Entanglement Frontier
1 mW Dilution fridge Cold stage
Commercial
Control
LBNL/ATAP
Control
3D Integrated Quantum Processor
Units (QPU)
Team: LBNL, UC Berkeley,
Bleximo, MIT-LL