Electronic-photonic arithmetic logic unit for high-speed computing

Zhoufeng Ying, Chenghao Feng, Zheng Zhao, Shounak Dhar, Hamed Dalir, Jiaqi Gu, Yue Cheng, Richard Soref, David Z. Pan, Ray T. ChenNature Communications 2020 · OpenReview

Abstract

The past two decades have witnessed the stagnation of the clock speed of microprocessors followed by the recent faltering of Moore’s law as nanofabrication technology approaches its unavoidable physical limit. Vigorous efforts from various research areas have been made to develop power-efficient and ultrafast computing machines in this post-Moore’s law era. With its unique capacity to integrate complex electro-optic circuits on a single chip, integrated photonics has revolutionized the interconnects and has shown its striking potential in optical computing. Here, we propose an electronic-photonic computing architecture for a wavelength division multiplexing-based electronic-photonic arithmetic logic unit, which disentangles the exponential relationship between power and clock rate, leading to an enhancement in computation speed and power efficiency as compared to the state-of-the-art transistors-based circuits. We experimentally demonstrate its practicality by implementing a 4-bit arithmetic logic unit consisting of 8 high-speed microdisk modulators and operating at 20 GHz. This approach paves the way to future power-saving and high-speed electronic-photonic computing circuits.

Introduction: Overcoming the Power Wall

For decades, microprocessor performance followed the trajectory of Moore’s Law, but modern computing has hit a significant "power wall." In traditional silicon transistors, power consumption increases exponentially with clock frequency ($P \propto f^3$). This physical constraint has led to the stagnation of clock speeds at a few GHz to prevent chips from overheating and failing.

A major bottleneck in arithmetic operations is the "carry" signal. In a standard addition operation, the carry from one bit must be calculated before the next bit can finish—a phenomenon known as the ripple-carry effect. While engineers use "Carry Select Adders" (CSAs) to pre-calculate results for both possible carry values (0 or 1), this requires doubling the hardware, which significantly increases space and power consumption. Researchers have now developed a hybrid electronic-photonic Arithmetic Logic Unit (ALU) that uses light to handle this "critical path" of mathematical operations.

The Electronic-Photonic Architecture

The core innovation of the Electronic-Photonic Arithmetic Logic Unit (EPALU) is moving the critical path—the part of the circuit where signals must wait for previous results—from the electrical domain to the optical domain. By leveraging the speed of light and the unique properties of photons, the system breaks the exponential power-frequency relationship found in pure electronics.

Overview of the electronic-to-photonic transition and system architecture.

Key technical features include:

  • Optical Carry Propagation Network (OCPN): Instead of electrons moving through transistors, photons travel through a network of microdisk modulators and passive directional couplers. Once modulated, light passes through stages using passive couplers that consume no dynamic power.
  • Wavelength Division Multiplexing (WDM): Instead of doubling hardware like an electronic CSA, the researchers used two different wavelengths of light ($\lambda_1$ and $\lambda_2$) to represent the carry-in states of 0 and 1 simultaneously within the same physical optical path. This acts as a form of "hardware compression."
  • Hybrid Integration: Electronics are retained for simple tasks like generating initial signals and calculating the final sum, while the high-speed carry propagation happens at the speed of light.

The Optical Carry Propagation Network (OCPN)

The OCPN is the heart of the EPALU. In this network, light travels through a micron-scale gate in sub-picoseconds, which is 1–2 orders of magnitude faster than electrons moving through a transistor with several fanouts. In these optical circuits, power consumption is proportional to frequency ($P \propto f$), not the cube of frequency ($f^3$). This "disentangles" speed from heat, allowing for much higher clock rates without the risk of thermal failure.

Detailed view of the m-bit Optical Carry Propagation Network (OCPN).

Experimental Results and Performance

The researchers evaluated their design using experimental measurements from a fabricated 4-bit OCPN chip and simulation benchmarks against the Synopsys Generic 32 nm design library. Key results include:

  • 20 GHz Operation: The team experimentally demonstrated a 4-bit ALU operating at 20 GHz.
  • Power Efficiency: At clock speeds above 20 GHz, the EPALU is projected to consume 1–2 orders of magnitude less power than state-of-the-art 7 nm transistor-based circuits.
  • Scalability: Simulations indicate that 32-bit and 64-bit architectures are feasible and could maintain these high speeds without the "heat death" associated with traditional electronics.

Limitations and Future Outlook

While promising, several challenges remain. As the bit-size increases, optical loss becomes a factor; light loses intensity as it passes through more couplers and modulators. For 64-bit ALUs, integrated optical amplifiers may be required, adding to the power budget. Additionally, the system currently requires static power for the laser source and thermal tuning to keep microdisk modulators aligned.

Future work will focus on monolithic integration of these photonic components with standard CMOS electronics on a single chip. The authors also suggest that future "plasmonic" modulators could shrink these components by 99%, potentially allowing for integration densities that rival modern transistors.

Summary updated

Abstract

The past two decades have witnessed the stagnation of the clock speed of microprocessors followed by the recent faltering of Moore’s law as nanofabrication technology approaches its unavoidable physical limit. Vigorous efforts from various research areas have been made to develop power-efficient and ultrafast computing machines in this post-Moore’s law era. With its unique capacity to integrate complex electro-optic circuits on a single chip, integrated photonics has revolutionized the interconnects and has shown its striking potential in optical computing. Here, we propose an electronic-photonic computing architecture for a wavelength division multiplexing-based electronic-photonic arithmetic logic unit, which disentangles the exponential relationship between power and clock rate, leading to an enhancement in computation speed and power efficiency as compared to the state-of-the-art transistors-based circuits. We experimentally demonstrate its practicality by implementing a 4-bit arithmetic logic unit consisting of 8 high-speed microdisk modulators and operating at 20 GHz. This approach paves the way to future power-saving and high-speed electronic-photonic computing circuits. Integrated photonics allows integration of complex optical circuits on a single chip. Here, the authors propose a wavelength division multiplexing based electronic-photonic arithmetic logic unit for computing at high speeds and with improved power consumption to help with the physical limits of Moore’s law.
BibTeX
@inproceedings{ying2020electronic,
  author = {Zhoufeng Ying and Chenghao Feng and Zheng Zhao and Shounak Dhar and Hamed Dalir and Jiaqi Gu and Yue Cheng and Richard Soref and David Z. Pan and Ray T. Chen},
  title = {{Electronic-photonic arithmetic logic unit for high-speed computing}},
  booktitle = {Nature Communications},
  year = {2020},
  url = {https://openreview.net/forum?id=oYwSm4M4l5},
}
Figure